The Inference Report

Pipeline figures in ML papers must be repurposed across many canvases, including paper columns, 16:9 slides, portrait posters, 1:1 social teasers, 9:16 phone previews. Each format imposes a different aspect ratio on the same computational graph, where any silently broken connection misrepresents the method. We formulate aspect-ratio-adaptive flowchart relayout as a distinct task: given a raster flowchart and a target ratio, produce a structurally faithful, hallucination-free, editable layout. Existing methods fail characteristically: image-to-image models stretch blocks and reject extreme ratios, text-to-image agentic systems hallucinate content, and parse-then-render systems mis-route edges. We propose an agentic pipeline factored into Parse, Style, and Layout stages, each pairing a main agent with a critic that combines deterministic constraint checks with VLM visual feedback so connectivity is explicitly checked and prevented from being silently broken. Outputs are draw.io-editable mxGraph XML. On a curated benchmark of 100 flowcharts at five aspect ratios, evaluated by Gemini 3.1 Pro and validated against human judgments, our method reaches 68.6% Content Fidelity versus 11.2-41.4% for prior work. Project page: https://onefigureeverycanvas.vercel.app/

In this paper, we study how training data creates associations between the tokens at the start of a base model's response and the reasoning behavior that follows. First, we demonstrate that fixing particular starting token cues makes a base model's performance competitive with that of its reinforcement learning (RL)-trained counterparts on math and coding. For instance, the cue ".\n\nOkay" raises Olmo-3-7B's MATH-500 pass@1 accuracy from 42% to 78%, while "Alright," raises Qwen3-14B's from 72% to 87%. Second, RL makes these cues more likely, while fixing them recovers much of its performance gain over the base model. Third, we trace the reasoning effects of token cues to the training data. We perform causal data interventions to turn an arbitrary word, such as "chicken", into an effective reasoning cue, or remove an existing cue's effect. A similar edit makes the prompt instruction "Think duck duck goose" as effective as "Think step by step" at eliciting reasoning. We also find that the hidden state representations induced by different cues correlate with different document types from the training set. Finally, we extend our study of token cues with a case study in language model safety, finding that different cues elicit distinct refusal and compliance behaviors that correspond to different types of training data.

Worst-group accuracy (WGA) evaluates a trained predictor but does not characterize how its frozen backbone behaves when a new head is learned. We introduce BiasFlow, a hook-based toolkit for monitoring class-attribute centroid alignment (IBMI), within-class centroid separation (W-IBMI), and feature-projection sensitivity. IBMI is confounded by class-attribute correlation and is not a measure of causal feature reliance. We pair these diagnostics with BiasFlow Regularization (BFR), a supervised, composable class-conditional centroid-alignment penalty. W-IBMI verifies the quantity BFR optimizes; it is scale dependent and does not independently establish attribute removal. Across the reported small-scale benchmarks, adding BFR improves or preserves mean WGA, with gains up to +26.0 pp on UrbanCars. The principal independent stress test freezes CelebA-Std backbones and trains fresh heads on biased data: BFR+GroupDRO improves WGA from 40.7% to 64.1%, while Male probe accuracy decreases from 92.5% to 72.2%. Attribute information remains recoverable, and cross-task results are mixed. A controlled synthetic-watermark ImageNet experiment additionally improves watermark-shift accuracy by +23.0 pp under matched training. These results support evaluating centroid geometry and resistance to biased head retraining alongside WGA, within the tested protocols.

Multimodal Diffusion Transformers (MM-DiTs) jointly process visual and textual representations throughout generation. These models repeatedly update the text tokens through multimodal attention, forming dynamic contextual tokens whose function is not well understood. In this work, we introduce a framework for reading this contextual space through natural-language interrogation. We train a lightweight bottleneck network that maps intermediate contextual tokens into the input space of a frozen Large Language Model (LLM), allowing the LLM to answer questions about the emerging image directly from these hidden representations. Our reader reveals that contextual tokens encode a rich, global representation of the emerging scene: generation-specific semantics, including attributes left underspecified by the prompt, are accessible surprisingly early in denoising, while increasingly fine-grained details become readable over time. Remarkably, this information remains decodable even when the MM-DiT receives an empty prompt, showing that contextual tokens accumulate substantial image-specific information from the evolving visual representation itself. We further find that generations with more readable contextual representations tend to receive higher human-preference scores. Building on these observations, we introduce Contextual Alignment, a training technique that explicitly reinforces the visual-semantic information encoded in the contextual tokens, improving generation quality and distributional coverage. Together, our results establish contextual tokens as both an interpretable view into the internal dynamics of MM-DiTs and an effective target for improving generative models.

LLM agents that orchestrate frozen vision-language-action (VLA) policies improve across episodes through text memory, which records what the agent did but not how the task is done. A demonstration video shows it, but fits poorly into an agent's context. The full video slows every turn, fixed keyframes lose the contact detail that decides whether a grasp holds, and what the agent needs shifts from the task's structure while planning to the frames around each contact. We introduce Recursive Video In-Context Learning (RV-ICL), a training-free method that turns a demonstration into a hierarchy the agent navigates rather than a prompt it receives. The hierarchy is built from the sub-events of the demonstration, such as grasps and releases. Its levels grow finer, from keyframes of the whole task to phases, moments and short clips, and are exposed through read-only tools. The agent reads the coarse levels before planning. During execution it re-enters the hierarchy whenever a step needs more detail and loads only the clip of its current sub-goal. One demonstration per task is enough. Built on RPent, RV-ICL raises success from 92.6% to 96.5% on LIBERO-PRO and from 86.7% to 95.8% on LIBERO-Plus.

Some diffusion posterior samplers construct Gaussian-tilted intermediate distributions along the reverse process. We observe that these targets can be pulled back to clean-space posteriors with weaker conditioning, with samples transported analytically to the corresponding noisy-space target through a Gaussian bridge. For the sequential Monte Carlo (SMC) sampler MCGDiff, the effective observation variance of this pulled-back problem is up to twice the diffusion-noise variance. We exploit this structure to initialize MCGDiff directly at an intermediate time: an approximate solver samples the softened clean-space posterior, the Gaussian bridge maps these samples to the tilted target, and only the remaining SMC suffix is run. This trades asymptotic consistency for finite-particle performance. With moment-matching posterior sampling (MMPS) as the solver, the hybrid improves sliced Wasserstein distance by roughly $2\times$ at matched particle count on a structured Gaussian-mixture inverse problem, and by more than an order of magnitude when the posterior-relevant mode is rare under the prior. A prior-initialization control, which retains the bridge but drops the clean-space conditioning, shows that on MCGDiff's standard Gaussian-mixture benchmark most of the improvement is insensitive to the conditioning. Conditioning the initialization gives a further consistent gain on the structured problem, and becomes decisive on a rare-mode problem, where resampling cannot repopulate a mode absent from the initial population.

Every recurrence of a looped language model adds cost in training, decoding, prefill, and reinforcement learning (RL). The closer recurrent states get to fixed points, the less the path to them matters. This enables truncated backpropagation in training; terminal key-value (KV) sharing for decoding with almost no loss in accuracy; a distilled student that prefills up to 1.79x faster; and RL updates that compute gradients from saved rollout states, 2x faster than backpropagating through the replayed trajectory. We therefore improve the two components of training that shape these fixed points: the depth prior and input injection. Fixed-depth training breaks KV sharing, and Huginn's broad depth prior supports sharing but dilutes supervision at the target depth more than sharing requires; we learn the prior from prediction feedback, with an entropy term that keeps it broad. Existing injection schemes let the state's component along the input amplify or cancel the injection; we remove this component with orthogonal injection. From 100M to 1.6B parameters, the learned prior and orthogonal injection lower perplexity at every scale relative to Huginn's prior and existing injection schemes, respectively. At 1.6B, the learned prior with a 3x smaller KV cache matches the downstream average of fixed-depth training with the full cache.

Sliders provide an intuitive interface for continuous image editing. In current generative approaches, however, the slider is simply a rescaling of the method's strength parameter, such as an adapter coefficient, a prompt weight, or an interpolation factor. This strength relates poorly to perceptual change. The image can partially revert as the slider moves, long stretches of the range produce no visible difference, and short intervals transform the image abruptly. Remapping the strength could fix this uneven pace, but only if the trajectory is monotone, which current methods do not enforce. We therefore distinguish the slider from the strength, and require perceptual distance from the input to grow linearly with the slider value. We introduce UniSlider, a lightweight LoRA trained on a few-step editing backbone so that its strength approximates this ideal slider. Few-step sampling lets us impose this objective in pixel space without intermediate ground truth, and the backbone's output is preserved at full strength. However, a low-rank adapter cannot make the strength fully uniform. Our slider is thus an inference-time remapping of the strength, obtained by adaptive sampling. Since training optmizes to make the trajectory monotone, this remapping closes the remaining gap without extra training or parameters. On a new benchmark of 300 continuous edits evaluating uniformity, monotonicity, edit fidelity, and identity preservation, UniSlider outperforms all prior methods and is preferred in a user study.

Memory has become integral to the LLM agent ecosystem, supporting information retention and reuse across interactions. However, most existing agent memory systems construct memory in a query-agnostic manner, which can incur unnecessary preprocessing cost and discard details that later prove essential. Recent studies have begun shifting memory processing toward runtime adaptation, but typically specialize in particular operations or fixed processing schemes, leaving flexible control over performance, cost, and latency largely underexplored. To address this challenge, we present \textbf{MemPilot}, a flexible framework that orchestrates on-demand memory curation under different performance--cost--latency preferences. Specifically, we optimize a multi-step LLM policy via reinforcement learning to iteratively choose between retrieving from query-agnostic memory and delegating query-specific curation of raw multimodal history to heterogeneous LLMs and VLMs. The policy jointly controls evidence amount, curation instructions, model selection, and visual access, enabling fine-grained allocation of runtime computation. To optimize this policy under competing objectives, we adapt objective-wise advantage decoupling by separately estimating each objective's advantage before aggregation. Moreover, we introduce prefix-based marginal utility estimation for fine-grained credit assignment across multi-step rollouts. Experiments on five multimodal agent-memory benchmarks demonstrate favorable performance--cost--latency trade-offs across optimization preferences, with preference sweeps yielding broader frontiers than existing trade-off-aware baselines.

Open-source web agents are now strong enough to execute realistic browser tasks, but training them with reinforcement learning still depends on weak supervision: binary task success is too sparse for credit assignment, while frontier-language-model judges are too expensive to call at every step and cannot be assumed available at deployment. We introduce CLIFT, a training and test-time scaling method built around conformal self-verification. During training, the agent answers natural-language verification questions about its own rollouts; a Compositional Conformal Certifier keeps only question signals whose URL-conditional evidence agrees with a training-time judge, assigns signed trust weights through polarity-aware lift, and blends the resulting verifier score into per-step rewards in a way that never subtracts from the judge baseline. At test time, the same certified bank is frozen and reused as structured evidence for Conformal Trajectory Selection (CTS): the agent samples a greedy rollout and one or more diverse retries, the self-verifier summarises each URL trace, and a conservative majority-vote rule chooses whether to swap away from the current incumbent without calling any external judge. This single mechanism supports three settings. On WebArena Infinity, CLIFT achieves state-of-the-art performance among open-source web agents. On VisualWebArena, a bank trained with the open model transfers to GPT-5.5 at test time and reaches state-of-the-art performance under the canonical harness. On Online Mind2Web, without training an agent on the benchmark, translating the certified question bank improves a live-web agent in zero-shot evaluation. Together these results position conformal self-verification as a way to turn costly judge feedback into a reusable training signal and a judge-free test-time scaling signal.

Scientific figures often encode quantitative results that are not readily available in machine-readable form, making accurate plot digitization important for verifying and reusing published findings. Yet it remains unclear how accurately current models recover plotted values from real scientific figures, as existing benchmarks rely largely on synthetic charts or cover only a limited range of chart types. We introduce PlotGround, an automated pipeline for building plot digitization benchmarks from real scientific figures and their author-released source data. PlotGround maps figures to source tables, identifies reconstructable panels, and generates quantitative questions with source-grounded reference values. We use PlotGround to construct PlotGround-1k, a human-verified benchmark of 1,119 questions from 1,066 bioRxiv preprints. Across sixteen multimodal models, the best reaches 87.5% accuracy at a $\pm 5\%$ relative-error tolerance. Tightening the tolerance to $\pm 2\%$ lowers every model's accuracy by 11-24 percentage points, revealing a gap between approximate visual reading and precise quantitative recovery. PlotGround's paired figure-source structure lets us compare how accurately the same values are recovered from figures and from source tables. Providing source tables instead of figures raises a coding agent's accuracy from 90.0% to 97.4% while cutting cost by 72%.

We introduce TasteVal, a benchmark to evaluate the experimental research taste of frontier models. We define research taste as the ability to pick interesting problems to solve, design experiments, and interpret experimental results. TasteVal measures the experimental component of research taste; given a fixed research problem, we measure how well a model iteratively designs experiments and draws conclusions from their outcomes. We operationalize experimental research taste as compute efficiency; a Researcher who reaches the same score as an expert human using half the serial experimental compute has twice the experimental taste. Experimental taste thus acts as a multiplier on experimental compute, making it a key input to forecasts of AI progress. TasteVal consists of 8 novel, challenging, open-ended tasks representative of frontier AI R&D. To isolate taste from coding ability, the model under evaluation acts as a Researcher that iteratively designs experiments while a fixed Coder agent implements them and reports their results. The Researcher executes until either the 40 H100 hour or 120 wall-clock hour budgets are exhausted. We recruit 24 human experts, at least 2 per task, and take the best expert attempt per task as the expert baseline. We evaluate 20 models released between 2023 and 2026. The best-performing model, Opus 5.5, exceeds our expert baseline, with a compute multiplier of 2.3x (95% CI 1.15-4.37), at roughly 1/30 of our baseliners' average per-run cost. On TasteVal, the compute multiplier of frontier models has doubled approximately every 3.0 months since December 2025 (95% CI 1.7-5.0), up from every 14 months between 2023 and December 2025. Measured by final normalized performance, frontier models show no trend break, doubling every 14.6 months. To keep TasteVal uncontaminated, we do not release the tasks.

Large longitudinal cohorts often contain wrist accelerometry without optical heart-rate sensing, motivating recovery of cardiac information from motion signals already collected during sleep. We present SeqSmoother, a transformer-based temporal corrector for sleep heart rate (HR) estimation from wrist accelerometry. SeqSmoother combines spectral descriptors with an intermediate Nightbeat-derived frequency anchor and a physics-motivated sub-harmonic feature designed to identify harmonic frequency lock-on. All inference-time features are derived from wrist accelerometry, while ECG is used only to construct reference HR labels and training-label quality weights. We evaluate SeqSmoother using 13 participant-disjoint held-out folds and compare it with the official Nightbeat implementation under a matched 60-s window and 15-s step protocol. Across all out-of-fold predictions, SeqSmoother achieved a participant-macro MAE of 1.60 bpm. On Nightbeat-retained matched intervals, Nightbeat achieved lower absolute error than SeqSmoother (0.615 versus 1.091 bpm), while SeqSmoother provided estimates over a larger portion of the eligible recording; Nightbeat produced final estimates for 72.85% of the SeqSmoother-eligible out-of-fold grid. Separately, the proposed sub-harmonic ratio achieved an AUROC of 0.972 for identifying reference-defined harmonic lock-on candidates. These findings reveal an accuracy-availability trade-off between learned temporal modeling and quality-gated signal processing while providing empirical support for a physics-informed approach to identifying frequency-tracking failures in accelerometer-based sleep HR estimation.

We study mistake bounds for differentially private online learning and online prediction under oblivious realisable adversaries. Online learning requires the learner to release a hypothesis at each time step whereas in online prediction, the learner only needs to make predictions without releasing a hypothesis. Using a novel lower bound for private online learning and an upper bound for private prediction, we show that the sample complexity of these two problems are separated by a factor that grows with the time horizon for every class of finite Littlestone dimension $d$. First, we prove that every $\br{ε,δ}$-private online learner has a deterministic realisable stream of length $T$ on which the mistake bound is at least $\bE\bs{M_T}=\Om{\frac dε\log\br{ T}^{2/3}}$. In particular, this is the first non-trivial lower in the range $1/T<δ<1/\log T)$ left open in earlier works[SR22,DSS24,LWY24]. Second, we prove that for every class of of Littlestone dimension $d$, there exists an $(ε,δ)$-jointly private predictor with at most $2^{2^{cd^2}}ε^{-2}\log^2\br{2/\br{εδ}}$ expected mistakes, independently of $T$, for some absolute constant $c>0$. Thus, for every fixed class of finite Littlestone dimension when $δ=Θ\br{1/\log T}$, private learning requires $\Om{\br{\log T}^{2/3}}$ expected mistakes, whereas private prediction admits $\bigO{\br{\log\log T}^2}$.

We distill Kokoro-82M, a widely used open text-to-speech model with 54 voices, into Paradee, an 8.07M-parameter model that speaks one of them. Paradee keeps Kokoro's architecture with much narrower layers, and each of its two halves is trained separately against the frozen teacher. It has 10x fewer parameters and needs 15x less compute. We first synthesize a corpus with the teacher and keep its durations, pitch, energy and phoneme features. We then train a small text side to predict these values, and a small decoder to turn the teacher's saved values into the teacher's audio, first with spectral losses and then adversarially. Finally, we connect the two halves and quantize the weights to int8. It needs no alignment learning and no joint training, and it runs on one laptop. Stored in int8, Paradee is 8.5 MB, runs 25x faster than real time on one CPU thread, and scores 4.41 on UTMOS against the teacher's 4.52. The student initially kept a slight buzz, which we trace to the phase of voiced speech between 2 and 8 kHz. A phase-locking filter applied after synthesis removes most of it, with no training and no extra parameters. Code, model files and audio samples are at https://github.com/sahilmahendrakar/paradee

World models learn to predict how their environment will evolve, making them an important foundation for general-purpose robotic control. Yet world action models depend on camera inputs whose manipulation can corrupt the visual representations used across tasks and action policies. Existing attacks on these models optimize against the victim's actions or predicted futures and therefore require access to target-model outputs. In this paper, we propose an attack, TAPDreamer, against world action models that instead uses a public encoder alone to construct a fixed local perturbation that transfers across tasks and action architectures. TAPDreamer requires no target-policy queries. Our key insight is that interactions between patch-induced changes in attention weights and value vectors broadcast a nearly identical representation shift far beyond the patch footprint, and this shift remains stable across task observations. Guided by this insight, TAPDreamer uses six frames from one source task to maximize the global L1 distance between clean and patched encoder representations. In closed-loop evaluation, one frozen patch per benchmark, covering about 6.5% of the input, reduces FastWAM's success rate from 97.7% to 0.0% across 40 LIBERO tasks and from 90.8% to 0.0% across 50 RoboTwin tasks; matched random patches retain 81.5% and 79.2% success. The same patches reduce success to 2.1% and 0.8% on two DreamWAM configurations and to 10.0% on Motus. These results show that protecting downstream action generation alone is insufficient: defenses for world action models must also secure shared visual encoders against persistent local perturbations.

We study agnostic tomography of pure bosonic Gaussian states: given copies of an arbitrary $n$-mode bosonic state $ρ$, the goal is to output a pure Gaussian state whose infidelity with $ρ$ is at most $\mathrm{opt} + ε$, where $\mathrm{opt}$ is the minimum infidelity achievable by any pure Gaussian state. We give efficient protocols achieving this in both the high and low fidelity regimes. When $\mathrm{opt}$ is below some universal constant, our protocol has runtime and copy complexity which is strongly polynomial in $n, 1/ε$ and $\log \log E$, where $E$ is the energy of the closest pure Gaussian state. For arbitrary $\mathrm{opt}$, our protocol uses $(n+1)^{\mathrm{poly}(1/ε)} \mathrm{poly}\left(1+\log\log(E)\right)$ copies and runtime. As a corollary, we obtain the first truly tolerant Gaussianity testing protocol for distinguishing whether $\mathrm{opt} > c + ε$ or $\mathrm{opt} < c - ε$, for any threshold $c\in(0,1)$. We also prove $\mathrm{poly}(n,1/ε)$ runtime is impossible, unless $\mathrm{NP}\subseteq\mathrm{BQP}$. Our protocols follow a shared paradigm: first, we iteratively use general Gaussian measurements combined with techniques from classical robust statistics to obtain a good warm start estimate, then we leverage non-Gaussian measurements to refine this warm start using convex and non-convex optimization methods. Interestingly, we prove that non-Gaussian measurements are necessary to match the strong agnostic guarantees we obtain, and in fact these guarantees are provably superior to what is possible for robustly estimating classical Gaussians.

Causal representation learning (CRL) is the process of recovering causally-related latent variables from high-dimensional observations. As a label-free inference method, CRL is particularly attractive for applications where data labels are unavailable or impractical to obtain. While there has been significant progress in understanding the identifiability guarantees of CRL, such guarantees often hold under highly stylized assumptions, which temper the direct application to real-world problems. This paper has a two-fold objective for interventional CRL. First, it establishes identifiability guarantees for substantially weaker interventional assumptions, resulting in block disentanglement of the causal variables, where the block structure depends on the realistically available intervention mechanisms. Secondly, the block disentanglement framework is used for embodied visual state estimation, in which the objective is to recover the latent physical variables of a robotic system directly from visual data (images and videos) without labeled data. These two components are critically complementary. The block disentanglement theory delineates identifiability guarantees under weakened assumptions, and the application demonstrates that the resulting objective remains effective in a controlled embodied setting despite further assumption violations, providing a theory-to-practice bridge needed to translate the promise of label-free CRL into practical problems.

Long-horizon planning with latent world models requires reasoning across timescales and levels of abstraction. Existing task-agnostic JEPA world models predict and plan at a single timescale or with multiple horizons in one shared latent space. We introduce H-JEPA, an end-to-end recipe for training a hierarchy of action-conditioned JEPAs in which each level predicts farther ahead in its own learned latent space. Planning proceeds top-down: the top level optimizes progress toward the goal, and each level's predictions become subgoals for the planner below it. When factors in the data evolve at separated timescales, higher levels discard fast, unpredictable detail and retain slower task-relevant state. Across four simulated navigation and manipulation environments, hierarchical planning improves over a flat JEPA; on Visual AntMaze, a three-level hierarchy raises success from 18% to 73% using less planner compute. Ablations attribute these gains to both temporal decomposition and higher-level goal representations. With inverse-dynamics supervision, the approach extends to diverse real-robot videos from DROID, where hierarchy improves offline planning fidelity at lower planner compute.

A language model can give a correct answer more probability than any single incorrect answer and still usually sample an incorrect one, because the incorrect answers together hold more probability. The power distribution raises each complete answer's probability to a power above one and renormalizes, shifting probability toward answers the model finds most likely (sharpening). Sampling from it improves reasoning without changing parameters, but needs many scored candidates per query. We show that a model can instead be trained to produce such answers in one generation. On-policy power distillation (OPPD) runs a sequential Monte Carlo sampler in which the model being trained generates candidates and a frozen teacher's power distribution weights them; the same probabilities weight each answer in a maximum-likelihood update. Training raises single-generation accuracy by up to 23.0 points on MATH500 and 27.3 on GSM8K over the untrained model at the same temperature, and one generation scores 2.4 and 3.5 points above published power sampling with 64 candidates, recovering 94 percent of the gain that 16 candidates give the untrained model. For context, against GRPO trained with verified rewards from the same checkpoint and budget, OPPD scores 3.8, 4.0 and 5.4 points higher on MATH500, GSM8K and AIME using no reference answers; the two are complementary, and OPPD applied after GRPO adds up to 9.3 points. Trained only on mathematics, OPPD raises HumanEval accuracy by up to 5.3 points. One loss coefficient moves the sharpening exponent the model absorbs between 1.19 and 2.02, against 1.14 for ordinary on-policy distillation, and it rises mostly on the model's own answers. Gains hold across model families and sizes, including a model already trained with verified rewards, where lowering the temperature gives nothing and OPPD adds 4.4 points on MATH500. Code: https://github.com/ArminAzizi98/OPPD.

Sparse attention is a primary approach to reducing the latency of diffusion transformers in long-sequence generation tasks, such as video and high-resolution 3D asset generation. However, existing methods can degrade generation quality and fidelity at high sparsity levels. Through controlled oracle comparisons, we trace this degradation to three sources: constraints imposed by token grouping, inaccurate interaction selection, and the attention contributions lost when tokens are discarded. Guided by this analysis, we propose Meta-Cached Sparse Attention (MC-Sparse), a training-free framework that selects individual key-value (KV) tokens while organizing similar queries into tile-aligned groups for efficient GPU execution. MC-Sparse caches metadata comprising query groups, KV indices selected using exact attention probabilities, and residuals between dense and sparse attention outputs, and reuses them across subsequent denoising steps. Across video and 3D generation models, MC-Sparse achieves higher fidelity to dense-attention outputs and larger denoising speedups than existing sparse-attention baselines, without visible quality degradation. Relative to dense attention, it delivers a $1.80\times$ denoising speedup on Minimax-H3-Base and a $2.32\times$ speedup on 3D asset generation, both with negligible quality loss.

Machine-learning emulators of chaotic and stochastic systems are usually validated on forecast skill and long-run statistics. Neither certifies that an emulator responds correctly to forcing, the property that projection and attribution studies rely on. Linear response theory makes this testable: the forced response follows from unperturbed correlations through a generalized fluctuation-dissipation relation, and decomposes over the stochastic Ruelle-Pollicott resonances of the Koopman generator. Building on the Koopmanism Response framework, we turn this into a calibrated, mode-resolved test for learned surrogates: each surrogate rollout passes or fails each check, and failure rates are compared with those of independent realizations of the true system. On stochastic Lorenz-63, a three-variable toy model, we evaluate SINDy, an MLP, a reservoir computer, a neural ODE and a neural SDE with learned diffusion, over up to 80 rollouts each. A sparse-regression model with the correct library passes every check at rates consistent with the true system. Invariant-statistics fidelity and response fidelity dissociate in both directions: a quarter of reservoir-computer rollouts pass every invariant-statistics check and match the static susceptibility $χ(0)$, yet misrepresent the slow relaxation modes, while the neural ODE and SDE rarely meet the invariant-statistics floor but recover those modes in three quarters of rollouts. As expected of a time-integrated quantity dominated here by fast relaxation, $χ(0)$ does not separate these cases. For a fixed network, the training formulation (one-step drift, flow map, or multi-step through the integrator) decides which of these properties it gets right.

We introduce Round-Trip KNN Clustering (RTKNNC), a graph-based method for finding cluster structure at several neighbourhood scales without requiring the number of clusters in advance. Unlike approaches that first make a $k$-nearest-neighbour (KNN) graph undirected, RTKNNC keeps both directions of the neighbour relation: which points a given point selects and which points select it. Incoming selections are treated as weighted votes that help decide which local connections remain visible during a recursive forward-and-reverse traversal. Repeating the procedure for increasing $K$ reveals how groups persist or merge as the neighbourhood scale grows; for the reference inverse-square model before structural refinement, clusters can merge but do not split. Because graph connectivity can occasionally join distinct groups through a sparse bridge or a small region of overlap, we add an optional label-free refinement. It first tests whether an already formed component is better described by two or three Gaussian subpopulations, and accepts a subdivision only when the proposed groups are large enough and consistent with the visible KNN graph. Across eight synthetic datasets and $K=2,\ldots,16$, independent C and Python implementations produced identical partitions in all 120 reference runs. Refinement increased adjusted Rand index from $0.7817$ to $0.9627$ on a variable-density benchmark and from $0.8083$ to $0.9853$ on a sparse-bridge benchmark. Comparisons with seven external clustering methods show competitive performance while preserving a label-free cluster-construction process.

The unprecedented computational scale of modern artificial intelligence depends on complex, multi-billion-transistor Systems-on-Chip, yet the workflows that verify these chips remain stubbornly manual. Although Large Language Models (LLMs) have made rapid inroads into Electronic Design Automation (EDA), approximately 74.6% of existing studies target static Register-Transfer Level (RTL) code generation, leaving post-simulation verification and interactive waveform debugging largely untouched. We introduce Back-to-the-Future (BTTF), an end-to-end agentic framework that closes this infrastructural gap. BTTF distills massive, unstructured simulation dumps into a normalized relational SQLite database and couples it with a collaborative multi-agent orchestration engine that translates natural-language verification queries into schema-aware SQL while correlating signal anomalies with versioned RTL repositories. Across a 150-query benchmark, BTTF attains 95.33% execution accuracy, charting a practical path toward autonomous EDA verification.

Much of the recent progress in machine learning domains such as language models has come from scaling laws that predict performance as a function of training effort. In high-energy physics (HEP) similar behavior has now been observed. To aid further study, we present a systematic procedure to derive robust scaling laws and compare design choices on the relevant budget axes for HEP tasks. We first validate the full scaling trajectory on toy problems and then apply the procedure to multi-task transformers on the ~11 billion-jet ATLAS JetSet2 dataset, in both the compute- and data-constrained regimes. For the latter, we predict, to the best of our knowledge for the first time, the jointly optimal model size, training horizon, learning rate and batch size under early stopping. At compute-optimal scaling, we recover a near-equal $\sqrt{C}$ dependence of model and dataset size, and find that auxiliary objectives lower the primary jet-classification loss at equal compute budget. Expanding the inputs toward lower-level data systematically lowers the loss while leaving the scaling exponent nearly unchanged. The onset of the power-law regime is itself set by scale: below a threshold in dataset size the loss carries little information about high-compute scaling, underscoring the value of large, high-quality full-simulation datasets as a foundation for scaling studies and the development of foundation models in HEP.

We present T-Search, an open-weight agentic retriever for hard multi-step search. Given a question and a search tool over a fixed corpus, it runs a bounded multi-round search and returns a ranked list of evidence chunks with short justifications, leaving answer generation to a downstream model, so backend and generator can be swapped without retraining. T-Search is built on Qwen3.6-35B-A3B and trained on adversarially filtered synthetic search tasks with round-sliced supervised fine-tuning followed by GSPO on a recall reward. Averaged over seven English and Russian benchmarks with gold evidence annotations, it reaches 56.0 Recall@10 with one rollout, 14.4 points above its base, and 61.3 with three fused rollouts, outperforming larger open models. We release the model, harness, live demo, and three benchmarks, including TRuST, the first native-Russian hard-search benchmark.

While modern AI detectors identify who wrote the words, emerging policies on AI use increasingly hinge on a different question: who came up with the ideas? We introduce IdeaLens, a detector that identifies whether a document's ideas came from a human or AI (idea provenance), regardless of who wrote its words. To focus IdeaLens on ideas rather than prose, we represent documents as outlines: lists of items that each pair a discourse role with a brief, paraphrased description of the content, minimizing word-level overlap with the raw text. We train IdeaLens on 1M FineWeb documents with silver labels from Pangram, a prose provenance detector. Since the outlines are largely stripped of surface-level information, the labels must be fit mainly through the ideas. In a controlled study, IdeaLens's AI flag rate drops from 95% to 7% as models write from increasingly detailed human plans, while Pangram 4 still flags 92%; from AI-derived plans, IdeaLens stays above 96%. Conversely, on a new dataset of 50 stories that human authors wrote from AI-generated plans, IdeaLens flags 68% of the stories as AI, compared to 8% for Pangram 4. On a comprehensive suite of 19 existing detection benchmarks, we show that IdeaLens maintains strong detection rates at low false positive rates, suggesting that ideas themselves provide a powerful discriminative signal, and its performance holds across domains, formats, and languages. Finally, we examine 90K predictions from IdeaLens to characterize systematic differences between human and AI ideation. We release our models and labeled datasets to facilitate future research on idea provenance detection.

The CADENCE algorithm solves the Partially Observable Cooperative Guard Art Gallery Problem (POCGAGP) with formal coverage and connectivity guarantees, but leaves unspecified which valid corner each agent should be deployed to, a choice that strongly affects efficiency. We introduce two learned corner-selection heuristics that preserve these guarantees: a CNN scoring candidates on a grid encoding, and a GATv2 network trained with Deep Q-Learning (DQN) on a visibility graph. Across 7,500 runs on random orthogonal environments (50x50 to 250x250), our heuristics outperform baseline CADENCE in both steps to full coverage and peak agent count, with gains growing with scale, and improve on Incremental Self-Deployment (ISDA) baselines in agent utilization while providing guarantees ISDA lacks. Learned corner selection thus improves CADENCE in speed and agent utilization at no cost to its formal properties.

Neural solvers for partial differential equations (PDEs) can approach an accurate solution while their parameters grow without bound. In such cases, the limiting solution may have no finite representation in the chosen model, leaving the best loss unattained. Our analysis connects missing limits in deep neural tanh- networks to unbounded hidden parameters or increasingly redundant neurons. For a class of models built from translated kernels, we describe the missing functions and recover them by adding kernel derivatives to the model. This completion makes the best approximation attainable under standard assumptions. Numerical studies follow the associated parameter growth and explore how completion affects PDE optimization.

Medical question answering spans specialties and clinical operations that may benefit from different adaptation directions. We propose ARBOR, a parameter-efficient method that selects rank-one components from a shared low-rank basis for each question. An additive gate combines question representations, specialty tags, operation tags, and their interaction; a learned coefficient scales the adapter residual. An illustrative separation under orthogonal, equiprobable subtasks shows how conditional selection can avoid an approximation floor faced by a fixed update with the same active rank. This result motivates the design without asserting a corresponding bound for medical corpora. On Qwen3-8B across CMB, CMExam, MedQA, and MedMCQA, five-seed experiments yield 69.69% mean accuracy across benchmarks, exceeding LoRA r16 and MoELoRA by 1.26 and 1.30 percentage points, respectively. The reported advantage over LoRA r16 increases from 0.08 to 1.94 points as training expands from one to seven specialties. Tag perturbations and atom masking support the usefulness of clinical routing, while atom clusters align with the supplied specialty labels (adjusted Rand index 0.62). Calibration, transfer, and measured costs further characterize the method. These findings support structured conditional adaptation for medical QA, while leaving clinical safety and broader deployment untested.

Matrix completion underlies problems from tabular imputation to causal inference, yet existing tabular foundation models treat it as entry-by-entry prediction, repeating context for every target and discarding the matrix's two-dimensional structure. We introduce MatrixFormer, a pre-trained matrix-native transformer that predicts a full distribution for every missing entry in a single forward pass. MatrixFormer is trained entirely on synthetic low-rank and latent-factor matrices under diverse missingness patterns. Applied zero-shot and with the same model weights, MatrixFormer achieves competitive performance on causal inference panel-data tasks, language-model benchmark-score completion, tabular imputation, and recommendation systems matrix completion. These results position MatrixFormer as a general-purpose foundation model for matrix completion.

Recurrent-attention hybrid language models (LMs), which interleave attention and recurrent layers, are increasingly used to combine the efficiency of the recurrent layers with the strong performance of attention layers. Prior work suggests that attention and recurrent layers offer complementary pathways to use past information: attention supports precise memory recall from earlier tokens, while recurrent layers support consolidation of disparate information over long contexts. However, we observe that simply having access to both pathways does not mean that hybrid LMs are effectively using them. We find that they rely substantially more on attention than on the recurrent state. Standard supervised fine-tuning improves overall performance but does not improve how the two memory pathways are coordinated: the model becomes more reliant on information propagated by attention layers, while its use of information propagated by recurrent layers remains limited. To encourage better coordination between the two memory pathways, we add an auxiliary loss that limits attention's access to earlier context while the recurrent state propagates through the full sequence. This objective encourages the model to retain and use information through the recurrent pathway alongside attention. It improves overall performance, with particularly strong gains on tasks involving longer contexts or requiring information aggregation, consistent with the strengths of recurrent layers observed in analysis. Crucially, this imbalance and the benefit of our auxiliary loss generalize: they apply to multiple recurrent-attention LMs in question-answering and agentic tasks, as well as to attention-based LMs that combine different forms of memory. Together, our findings show that simply providing multiple memory pathways does not ensure their effective use, and that targeted supervision is needed to better coordinate them.

In decentralized consumer-to-consumer (C2C) marketplaces, people list goods, negotiate with strangers, and rate one another, so trust rests on reputation. Large language model (LLM) agents now act for users, raising risks to their money, privacy, and reputation. We introduce BazaarBench, a simulated C2C marketplace and benchmark for evaluating the safety of these agents. It tracks ownership, item condition, and commitments across transactions, combining record checks with rubric-based LLM judgments to identify six failure types across five stages. We run three base markets for 30 simulated days, each with 100 agents using one model and inventories drawn from a public eBay sample. Across 45 continuations, we evaluate five models under ordinary instructions, deadline pressure, or adversarial instructions to exploit other traders. Each continuation runs for seven simulated days from a copy of a market's day-30 state. The tested model controls the same 20 selected agents, retaining their personas, inventories, and histories, while the other 80 keep the base model. All five models attempt to promise the same item to multiple buyers under ordinary instructions. Adding targets and deadlines increases these attempts for every model. Under adversarial instructions, the share of tested sellers' committed transactions completed despite unavailable items or overstated conditions rises from 15.4% to 33.4%, reaching 55.5% for GPT-5.4. Averaged across models and markets, simulated weekly earnings per tested agent rise from USD 20 under ordinary instructions to USD 33 under adversarial instructions. Most of the increase comes from items the sellers never held. We release the simulator, saved market states, evaluation code, and records covering 357,608 agent model calls for evaluating new models and developing safer marketplace agents.

Hierarchical graphs embed in hyperbolic space with lower distortion than in Euclidean space owing to its negative curvature. However, their gradient-based learning is hampered at large radii, where the Poincaré ball and the Lorentz hyperboloid models fail numerically. Polar coordinates avoid this problem, but the hyperbolic metric scales the angular step by the hyperbolic sine of the radius, freezing angular motion. We observe that this factor is a choice, silently fixed by existing implementations: the Euclidean tangent parametrization, for instance, uses the radius itself. We show that other choices are not only possible but preferable. They are endpoints of a one-parameter family of optimization preconditioners with curvatures from $-1$ to $0$, while the embedding remains at curvature $-1$. We show that since the Euclidean preconditioner rearranges a layout but refines it poorly, while an intermediate one refines far better once a layout is in place, combining them in two stages reduces the loss on real-world trees by 46-74% over the best single curvature.

ufakzeka-karar is an open Turkish decision model with 182,494,466 parameters. Given a Turkish text and questions of a fixed answer type (a choice, a level on an ordered scale, or yes or no), it returns a temperature-scaled probability for every option and an expected error that serves as a "not sure" signal, without generating text and in one CPU forward pass for up to ten options. Built on the lab's ufakzeka-1-base, its head scores each option blind to the others at shared positions, so the answer does not depend on option order. A sequential head trained with shuffled options was about as accurate but changed 2.3 to 2.8 percent of its answers when only the option order changed; REINFORCE lost 10.2 points (0.102) of macro F1 to cross-entropy. On the open set of HakemBench v1.0 (4,275 questions, 7 tracks) the released model ranks 7th of 16 rows with a composite of 0.660 (95% interval 0.642 to 0.677). Temperature scaling lowers calibration error (smooth ECE) on the development set but raises it on held-out support questions, from 0.027 to 0.045 for the first scored run, which never trained on them; the released model later trained on them, so its 0.036 to 0.064 is not an unseen-question test. The released model is the last of three runs scored on HakemBench, and its numbers are not blind. The second run's new training data was aimed at the first run's errors on the full test set in guardrails, moderation and customer support, and the released run was trained after the second run's guardrail results on the full test set were read, under a protocol fixed in writing before any of its data, code or runs. All its numbers come after these readings; its guardrail, moderation and customer support numbers carry the flag "shaped by reading the test results". With every model scored on the other four tracks only, its composite is 0.678, 6th of 16. Weights and code are under Apache-2.0.

Large language models (LLMs) can generate accurate responses, but these are void of diversity. We attempt to address this for the task of creative short story generation. Drawing on established writing conventions and known LLM limitations, we target variation in genre, tone, style, and named entities. To promote diversity across these dimensions, we introduce DivLM, an LLM post-training framework consisting of two phases. First, we perform continued pre-training on a creative writing corpus and restore instruction-following capabilities using weight residuals. We then apply reinforcement learning with a custom, composite reward function that jointly maximizes diversity across the targeted narrative dimensions while maintaining response quality. Our empirical results on two LLM families show that DivLM increases diversity metrics by more than 9% on average compared to alternative approaches, while preserving instruction following, overall response quality, and similarity to human outputs.

Electroencephalography (EEG) offers millisecond temporal resolution, but inferring underlying neural sources is a severely ill-posed spatial inverse problem. While deep learning has advanced spatial reconstruction, current architectures face a critical dilemma: frame-by-frame models discard vital temporal context, whereas full 4D spatiotemporal networks introduce an architectural trade-off between reconstruction accuracy and inference cost. We propose a novel two-stream framework that explicitly decouples global temporal representation learning from per-time-point spatial refinement. A Transformer-based Temporal Condition Encoder processes the entire EEG sequence via factorized spatiotemporal attention, retaining sensor-resolved features. A fixed inverse then maps these features into source-indexed conditioning for a per-timestep Source-Space Transformer or volumetric convolutional refiner. Extensive evaluations on realistic synthetic data demonstrate that this temporal prior dramatically improves spatial localization, outperforming classical and spatiotemporal baselines, particularly in high-noise and multi-source regimes. Training across diverse leadfields and explicit operator mismatches improves transfer to unseen head geometries and brings template-based reconstruction closer to subject-specific inversion. Furthermore, we apply the model trained only on synthetic EEG data to real-world EEG. A logistic regressor fit on source power differences in eyes-open, eyes-closed conditions successfully decodes age groups.

Mixture-of-experts (MoE) layers increase model capacity without a proportional increase in per-example computation. However, conventional flat routers can yield imbalanced expert utilization and treat experts as an unstructured collection, whose indices carry no topological meaning. We introduce {\bf BRANCH-MoE}, a routing architecture that places \(E\) experts at the leaves of a binary decision tree of depth \(\log_2 E\). At each internal node the branching probability is centered on the arrival-weighted mean score of the traffic reaching that node. This mean is estimated using an exponential moving average, which promotes utilization of both child subtrees without an auxiliary load-balancing loss. We show that this moving-average estimate admits an explicit noise-lag trade-off. We prove that for linear node maps and log-concave arrival distributions, this mechanism prevents routing-mass collapse. We further establish that, under a frozen router, an expert's execution frequency controls its stochastic-gradient convergence rate, and that confident decisions near the root bound cross-device communication when experts are assigned to devices by tree prefix. We evaluate BRANCH-MoE against Switch softmax, DeepSeek-V3 dynamic-bias, Skywork logit-normalized, and deterministic hash routing on Criteo click-through-rate prediction, Forest Covertype, HIGGS, and YearPredictionMSD, using \(E=16\), top-\(4\) routing, and five random seeds. Our results show that hierarchical routing can preserve task quality and balanced utilization while inducing a topology that supports localized expert co-activation and reduced communication.

We investigate whether continued pretraining on Russian legislative documents improves a Russian ModernBERT encoder on legal text. The adapted model, RuModernBERT-ruLaw, was trained on a corpus reported to contain 304,382 legislative documents and 194,425,905 corpus tokens. Corpus token counts are distinguished from positions produced by the model tokenizer. We compare the original and adapted encoders on a fixed external collection of 1,031 court-decision segments. Both models receive the same hidden positions in each of five masking realizations. At maximum input lengths of 512, 2,048, and 8,192 tokens, mean masked-token cross-entropy decreases by 0.10942, 0.07052, and 0.06604 natural-log units, respectively. The reported 95% intervals summarize sensitivity to masking on this fixed collection; they do not quantify uncertainty across document collections. A second evaluation addresses legal-entity extraction. The original and adapted models achieve entity-level F1 scores of 0.99852 and 0.99820. However, 99.95% of test spans have the same normalized surface form and class in the training split. This evaluation therefore provides limited evidence about transfer to previously unseen forms. The paper explains the masking objective, overlapping windows, averaging rules, and exact entity-boundary scoring using editable diagrams and clearly marked illustrative examples. The comparison supports lower masked-token prediction loss for the studied pair of models and collection. It does not isolate the contribution of distant context or establish practical legal utility.

Learning the Hamiltonian of a many-body system from its dynamics is a central task in quantum science, yet the algorithms with the strongest provable guarantees assume some level of quantum control--fast, arbitrary single-qubit gates interleaved with time evolution, and measurements in arbitrary bases--that is beyond the capabilities of near-term analog quantum simulators. Motivated by analog atom- and ion-based platforms, we study Hamiltonian learning under minimal access models. Uniform state preparation and measurements: We first consider the setting where in every experiment, one can rotate each qubit to the same state, perform short-time evolution, and measure every qubit in the same basis. Surprisingly, we show that for generic 2-local Hamiltonians on any interaction graph, all of the parameters can be reconstructed from such experiments. Computational basis state preparation and measurements: We then consider a similarly constrained setting, but where state preparation and measurement are restricted to the computational basis. For nearest-neighbor Hamiltonians with only Pauli $X/Z$ interactions, a class which captures contemporary Rydberg atom platforms, we show that over 1D and 2D rectangular lattices, all of the parameters can be reconstructed from such experiments up to unavoidable gauges. Our protocols introduce new techniques for solving structured polynomial systems over an extensive number of parameters. Taken together, our results suggest that one can learn a great deal from the dynamics of quantum many-body systems even under the most stringent experimental constraints.

Multimodal rumor detectors increasingly rely on retrieved evidence, yet relevant evidence is not necessarily sufficient for verification. Missing provenance, duplicated reports, and unresolved contradictions can produce confident predictions without adequate support. We introduce SAFE-MR, a framework that separates claim veracity from evidence sufficiency. The method decomposes image-text posts into verifiable claims, constructs a relation-aware claim-evidence graph, and aggregates evidence using provenance and contextual compatibility. Separate veracity and sufficiency heads support selective prediction, while evidence interventions encourage stability under irrelevant additions and sensitivity to evidence removal. On NewsCLIPpings, VERITE, and XFacta, SAFE-MR achieves macro-F1 scores of 91.2%, 75.8%, and 85.2%, respectively. Against the matched backbone with evidence, its macro-F1 gains are 2.2, 4.9, and 4.8 percentage points. On the diagnostic selection set, SAFE-MR reduces AURC from 0.105 for maximum-probability rejection to 0.075 and lowers error at 80% coverage from 13.8% to 8.5%. Evidence-perturbation and ablation results support the role of sufficiency learning and intervention training in improving selective verification.

Background music that matches the mood of a text has been shown to make readers feel more immersed and improve their reading experience, motivating recommender systems that pair books with mood-matched music. In this direction, we present Sentiment Aware Generative Adversarial Network for Cross Domain Recommendation (SAGA-CDR), a two-phase cross-domain recommendation framework that personalizes music suggestions and emotionally aligns them with the book being read. In the first phase, transformer-based sentiment embeddings are constructed from user reviews and mapped across domains via a Conditional Generative Adversarial Network, whose mask-conditioned generator handles missing sentiment components and injects stochasticity for richer preference transfer. A compact rating neural network then fuses sentiment-specific interaction scores with a collaborative filtering prior to predict music ratings. In the second phase, large language models classify each book into a valence-arousal emotional quadrant, and candidate tracks are filtered to match that quadrant. Experiments on both the English Amazon and Chinese Douban datasets show that SAGA-CDR achieves the best rating prediction accuracy on Amazon (RMSE 0.98) and the lowest RMSE on Douban (0.91), with ranking performance competitive with the strongest sentiment-aware baseline, even in cross-lingual settings.

A recurring design principle in modern optimizers is to decouple update magnitude from the raw gradient norm, yet its consequences for learning-curve and resource scaling remain unclear. We isolate this mechanism by studying normalized SGD in a random-feature model with power-law teacher and data covariance. Fixed-norm updates induce an effective learning rate that grows as gradients shrink. We derive a dynamical mean-field theory (DMFT) describing the joint dependence of the loss on training time, model width and batch size. Normalization initially accelerates SGD, mapping the power-law exponent $r_{\rm SGD}<1$ to $2r_{\rm SGD}/(1-r_{\rm SGD})$, with exponential convergence at $r_{\rm SGD}=1$ and formal finite-time convergence for $r_{\rm SGD}>1$. At finite step size, however, the same feedback ultimately breaks the acceleration and leads to marginal stability. The late-time theory yields width-limited, edge-of-stochastic-stability (EoSS), and deterministic edge-of-stability (EoS) regimes. These phases determine when larger batches or wider models reduce serial training time at comparable compute. We quantify in which of these phases increased batch size or width can compensate the excess compute use per step by fewer optimization steps to target loss. Linearized ResNet experiments on CIFAR-5M support the predicted acceleration, breakdown, and resource-scaling trends. Together, these results connect normalization-induced acceleration, EoS effects, and width--batch allocation within a solvable theory.

While medical multimodal large language models (Med-MLLMs) advance medical visual question answering (VQA), existing clinical workflow-inspired multi-agent frameworks suffer from interaction patterns and excessive computational overhead caused by redundant communication topologies. In this paper, we propose MedPrune, an efficient medical multimodal multi-agent collaboration framework that dynamically prunes both nodes and edges from the communication topology to enhance reasoning ability and token efficiency. Specifically, we first formulate the diagnostic process as a heterogeneous communication graph, where nodes represent specialist agents from various departments and edges capture intra- and inter-departmental interactions. Building on this graph, we introduce two sparsification mechanisms to enable adaptive collaborative evolution: (1) Heterogeneous Node Sparsification, which eliminates task-irrelevant specialist agents irrelevant to the current multimodal question via reinforcement learning-driven topological optimization, and (2) Heterogeneous Edge Sparsification, which selectively retains only the most diagnostically salient intra- and inter-departmental connections by jointly optimizing task performance and topological complexity. Extensive medical VQA experiments under full-set and few-shot training settings prove MedPrune surpasses multi-agent baselines and boosts token efficiency with strong adversarial robustness.

Graph foundation models aim to transfer across graphs, feature spaces, relational schemas, and prediction tasks, yet existing approaches typically generalize only within particular graph modalities or tasks. We propose Wander, a graph foundation model designed to operate across these settings within a single pretrained checkpoint. Following the prior-predictive perspective, we formulate graph learning as completion of a partially observed graph. We realize this task-general view through a common interface based on random walks, allowing the same model to operate across homogeneous and multi-relational graphs with varying features, labels, and relational schemas. Wander can increase its structural context at inference time without changing its learned parameters and, under suitable assumptions, universally approximates the corresponding Bayes-optimal predictor on bounded connected graphs. Empirically, a single pretrained checkpoint achieves state-of-the-art or highly competitive results across node classification, homogeneous link prediction, and knowledge-graph link prediction. Moreover, joint pretraining across graph modalities and tasks preserves performance in specialized settings while enabling positive transfer and the composition of separately learned capabilities at inference time.

While deep features have transformed anomaly detection in images and video, their impact on tabular data has been less substantial, partly due to the limited availability of strong deep representations. Recently, prior-data fitted networks (PFNs) have emerged as a promising source of such representations for tabular data. In this work, we investigate how PFN representations can be adapted and leveraged for anomaly detection. The question is harder than it looks. No anomalies are available before deploy- ment, so model parameters cannot be tuned with supervision, and the reference set that defines normal behavior may itself contain the very anomalies it is supposed to reveal. We begin our study using frozen TabPFN features. Scoring each sam- ple by its distance to its nearest neighbors in feature space already gives strong results. We identify which layers to use and a feature-extraction procedure suited to the task. Next, to further improve performance, we use the reference set to fine- tune the model, so that the resulting features better separate normal samples from anomalies. On the ADBench benchmark, our fine-tuning free approach (ZEN) reaches a higher mean AUROC than every baseline, and our fine-tuned method (FOCUS) improves on it further. Our approach also generalizes across PFN models.

Embedding enhancement improves speaker verification under acoustic mismatch without modifying a frozen backbone. Recent work has established a practical label-free setting for this task, but often adopts increasingly structured formulations. Here, the clean target is directly observed during training, making enhancement a matching problem on the unit hypersphere. We model the clean target with a von Mises--Fisher (vMF) likelihood and profile out a sample-wise concentration parameter, yielding a simple closed-form objective with adaptive weighting. Across VoxCeleb1, VoxSRC23, CN-Celeb, VOiCES, and VC-Mix, the proposed method largely preserves the baseline and gives clearer gains on challenging mismatch sets. It also remains stable under a broad single-view recipe, where a recent diffusion baseline becomes less reliable in controlled comparisons. These results suggest that effective label-free embedding enhancement in this setting does not require a highly structured formulation.

Search agents adapt their queries, yet fixed search interfaces leave candidate processing and evidence presentation outside the agent's direct control. Our trajectory analysis shows that supporting passages can be retrieved yet never delivered to the agent; a same-page oracle intervention shows that changing the returned evidence can reduce subsequent search. We introduce Programmatic Search Agent (PSA), which makes a local executable computation over candidates the unit of a search action. PSA unifies a persistent candidate workspace, flexible primitive composition, and selective evidence presentation. It incrementally generates program cells that reuse candidates, execute dependent operations, and select what the agent inspects next. The runtime resolves specified data dependencies within each cell, while the agent adapts its search strategy across cells as new evidence arrives. We compare PSA with the Query-based Agent and Tool-based Agent on InfoSeek-Eval and BrowseComp-Plus using five policy backbones without task-specific training. All three interfaces share the search substrate, and the Tool-based Agent also shares PSA's primitives and persistent workspace. Relative to the Query-based Agent, PSA improves macro-averaged task success by 4.00 and 7.56 percentage points on the two benchmarks, respectively; within-backbone reductions in final-step tokens average 28.3% and 33.9%. These results support extending agent control beyond query reformulation to the processing and presentation of retrieved evidence. Code will be released subject to approval.

Long context inference with large language models becomes increasingly expensive as attention must operate over an ever growing KV cache. Page sparse attention reduces this cost by representing each KV page compactly and retrieving only a subset for each query. Existing retrieval methods are designed to estimate attention scores or page relevance, but their objectives do not directly account for how approximation errors affect the resulting value weighted attention output. We introduce \method{}, an output aware page encoding derived from the joint structure of keys and values while preserving the key information needed for accurate retrieval. \method{} is training free and requires no additional value dependent statistics at inference time. Once constructed, its stored representation has the same size and decode time scoring cost as a key only spectral representation. Across long reasoning, long context understanding, and long generation benchmarks, \method{} consistently improves over the key only spectral baseline and performs competitively with recent KV cache compression and retrieval methods. On long reasoning benchmarks, it achieves strong avg@\(k\) performance across model benchmark pairs, while matching or surpassing leading baselines on several long context understanding and generation settings with modest decoding overhead. Code is available at \url{https://github.com/Ashkan13776/oval-kv}.

Clinical questions often depend on linking a patient's multimodal evidence to external biomedical knowledge, yet existing predictive systems rarely represent such links explicitly, so they can neither be traced to their evidence sources nor removed to measure their contributions. We present MM-KG (Multimodal Knowledge Graph), which represents heterogeneous, multimodal patient observations and biomedical concepts as separate layers in one typed graph, joined by explicit alignment edges. First, modality-specific harmonizers convert EHR text, imaging, genomic, and biospecimen data into typed observations mapped to UMLS concepts, which a route-prioritized aligner links to a biomedical knowledge graph. Query-conditioned retrieval then selects a compact subgraph for downstream use by a large language model or a graph neural network. We build MM-KGs for MIMIC-IV and ADNI, and evaluate them with a 2x2 design that separates patient evidence, biomedical knowledge, and their interaction. On questions that require both sources, neither source alone performs far above chance, whereas their combination yields a drug-controlled AUROC interaction of +0.194 on MIMIC and +0.299 on ADNI. On held-out five-candidate ranking, MM-KG outperforms MindMap by +0.131 Hits@1 and leads an adapted GraphCare on the items that require consulting the patient, and deleting the single answer-bearing relation from the retrieved packet returns Hits@1 to the no-knowledge baseline. Finally, query-conditioned retrieval reaches 0.731 AUROC with 6.8x less context than the strongest generic policy, whereas static knowledge graph context gives no consistent gain on ordinary outcome prediction. Knowledge graphs thus benefit clinical LLMs not as background context but as explicit links between multimodal patient evidence and the relation a question requires, and MM-KG makes these links retrievable, traceable, and testable.

Tabular foundation models perform in-context learning (ICL) by conditioning predictions on labeled training examples provided as context. Unlike traditional models that separate training from inference, these models must process all training examples in every forward pass, making each prediction expensive. Restricting the number of training examples reduces this cost but substantially degrades performance. Instead of discarding context, we propose activation alignment, a method that leverages the full context to teach a model how to behave when seeing only a subset. This is achieved by training a lightweight linear transformation on synthetic unlabeled data to map the intermediate activations of a data-constrained "student" (using partial context) toward those of a full-context "teacher" (using all data). Training the aligner requires no GPU and converges in seconds to minutes on commodity hardware. We evaluate on 38 classification datasets from the TabArena benchmark using the leading two tabular foundation models, TabPFN-3 and TabFM. Across all context budgets, the aligned student yields broad, statistically significant improvements over the unaligned baseline for both models. In low-data regimes, alignment recovers nearly half of the teacher's predictive advantage. The method provides a practical, low-overhead approach to achieving the inference speed of compact contexts while closing a significant fraction of the performance gap to the full-context teacher.

Mixture-of-experts (MoE) routers typically apply softmax to the router scores and keep the top-K experts, making every token use exactly K experts. Sparsity-inducing probability maps such as sparsemax, alpha-entmax and normmax can adaptively assign exact zeros to selected experts, and therefore appear to offer token-dependent expert participation, even when using the same top-K machinery. In this work, we study whether and how this sparsity survives training. We train matched 300M and 1B top-2 MoE language models with softmax, 1.5-entmax, sparsemax and 2-normmax, and find that the maps behave very differently once trained: at 1B, entmax discards 30% less probability mass than softmax while almost never dropping a selected expert, sparsemax retains the most mass, and normmax routes 21% of tokens to a single expert. These outcomes are not properties of the maps alone. Each map drops a selected expert only when the gap between the two largest scores reaches a fixed threshold, and the trained routers differ in the score distribution they learn: the entmax router learns scores with roughly half the spread of softmax's, which keeps its top-2 gaps below its threshold, while sparsemax and normmax, which share the same threshold, learn different gap distributions and hence different participation. Routers thus co-adapt their scores to the map, and a map's capacity to produce zeros does not by itself determine expert participation. While none of the sparse maps improves validation loss over softmax, they make the trained models far less sensitive to selecting more experts at inference: sparsemax trained with K=2 loses 0.02 nats when run with K=8, where softmax loses 0.58. Our results indicate that adaptive MoE routing has to be designed around the joint behavior of the probability map and the learned scores, rather than around the map alone.

We study gradient descent (GD) with a large constant stepsize for logistic regression on linearly separable data. Existing analysis shows an accelerated rate of $\widetilde{O}(1/\sqrtε)$ to reach loss $ε$ with an aggressive stepsize, although the loss may initially oscillate. Tighter control of the oscillatory dynamics has been available only for two-dimensional data. We prove a substantially faster rate in arbitrary dimension: GD with a large stepsize $η=1/ε$ reaches loss $ε$ within $O(\ln^{p}(1/ε))$ steps, where $p$ depends only on the margin and the rank of the data. Our proof improves the bound on the transition time of GD from the oscillatory to the stable phase, after which the loss decreases monotonically. We split the oscillatory phase into recursively nested intervals. The margin and the rank bound the nesting depth, and a counting argument bounds the number of intervals at each depth, together yielding the polylogarithmic step complexity.

Nighttime lights from NASA's Black Marble product suite capture thermal and light emission signals from anomalous events including fires, volcanic eruptions, and gas flaring. Existing detection approaches rely primarily on thermal bands, limiting sensitivity to weaker signals. We propose a novel machine learning framework that jointly models Black Marble M-band and Day/Night Band (DNB) signals to derive a generalized, spatio-temporally robust ensemble of anomaly detectors. The framework iteratively builds detectors that scale across regions, seasons, anomaly classes, and extends over land and ocean. Detection sets at varying confidence levels are derived based on relevant bands and detector agreement. The approach improves true detection rate while reducing spurious detections and results demonstrate strong generalizability with applications in natural hazard monitoring and energy extraction.

Are language models compliant with user instructions? A model that always complies can be stopped but also exploited, while one that always resists can be neither exploited nor stopped. We contribute an open two-probe benchmark that can place any language model on this spectrum. In the active probe, a user instructs the model to act and accept a lower payoff, which measures exploitability. In the passive probe, the user instructs it to wait and give up a higher payoff, which measures stoppability. The two compliance rates combine into a compliance index $κ$. Applied to twelve language models, the benchmark shows that seven mostly follow the instruction in both probes and justify their action by pointing to the instruction. Only Claude Sonnet-4.6 and Claude Opus-4.7 can be stopped without being exploitable, Claude Opus-4.6 and GPT-5-mini resist both instructions, and no model is exploitable but unstoppable. Knowing where a language model sits on the compliance index $κ$ matters for human operators and for multi-agent systems, whether distributed or orchestrated.

Long-video understanding places substantial demands on memory, as answering questions often requires retrieving information distributed across extended temporal spans. Existing approaches broadly follow two paradigms: query-driven exploration, which is sensitive to localization errors, and query-independent memory construction, which may omit question-specific details. We introduce VideoTapestry, a training-free multi-agent framework that adapts a preconstructed hierarchical video memory through coarse-to-fine, query-driven refinement. The preconstructed memory organizes video content into three levels, capturing global narrative context, event-level temporal structure, and fine-grained relational evidence, respectively. To support coarse-to-fine localization and observation, we assign a specialized agent to each level, keeping retrieval and refinement within a scale-specific context. Guided by the query, these agents revisit relevant video regions and enrich layer-wise memories with targeted multimodal observations. Their refinements are assembled according to the original hierarchy into a composite query-adaptive memory, preserving global context in a compact form while retaining fine-grained evidence along query-relevant branches for final reasoning. Compared with direct GPT-5.5 inference, VideoTapestry achieves absolute accuracy gains of 17.2%, 14.9%, 9.8%, and 7.0% on LVBench, LongVideoBench (Long), Video-MME (Long), and EgoSchema, respectively, achieving the state-of-the-art results among all competitors.

Small language models are often post-trained as students on reasoning traces from stronger teacher models to efficiently learn new skills. However, token-level imitation on traces that lie far outside the student's expected distribution often produces \textit{confident conflicts}, whereby the student is required to imitate a continuation that it deems unlikely (i.e., low-probability) despite being confident in a different continuation (i.e., in a low-entropy state). To mitigate the degradation in generalisation and catastrophic forgetting caused by these conflicts, we propose \textbf{Entropy-Aware Mixing}: a dynamic per-token interpolation of the student and teacher distributions, gated by the student's predictive entropy. We implement both convex and geometric interpolations for both offline trace generation (via speculative decoding, then SFT) and on-policy forward-KL distillation. Our results show that entropy-aware mixing stabilises distillation, improving in-distribution and out-of-distribution math reasoning while better preserving general capabilities than fixed-teacher supervision. Nonetheless, the optimal entropy schedule depends on the training source, with offline-generated traces favouring concave schedules (greater overall teacher influence) and on-policy training favouring linear or convex schedules (teacher concentrated in high-entropy states).

Adversarial attacks on Large Language Models (LLMs) aim to induce harmful content. However, existing methods suffer from high computational costs or strict model-pairing dependencies, limiting their scalability and transferability. We propose Reward Stealing Attack (ReSA), an adversarial attack framework that targets the latent safety reward underlying LLM alignment. ReSA employs maximum entropy inverse reinforcement learning to recover a proxy reward model solely from the aligned model's behavior. The extracted reward is then reversed at inference time to derive an adversarial policy, efficiently implemented via a reward-guided decoding mechanism. Experiments demonstrate that a single recovered reward generalizes across prompts and diverse models to reveal a fundamental alignment vulnerability, enabling ReSA to significantly outperform existing attacks in effectiveness and transferability. The code is available at https://github.com/GarminQ/ReSA.

Language models draft engineering calculations, but answer accuracy does not show whether they reject an impossible problem. We tested 14 models on 30 pairs of mechanics problems, each with a valid version and one made impossible by changing a given value or assumption. Two independent solvers verified every answer key and showed that each flawed problem was physically impossible. We scored solving of valid problems separately from rejection of their flawed counterparts. Each reply required a "solved" or "cannot solve" status; rejection meant "cannot solve" or withholding an answer. The initial prompts did not warn that problems could be flawed. Across three recent models, 12 of 90 replies failed to reject a flawed problem. In 11 of these replies, the model stated the flaw, answered a corrected problem and still reported the original as "solved", according to artificial intelligence raters and numerical checks. We later retested four models from one provider, offering "flawed" instead of "cannot solve" and asking them to name and explain the defect. Three models showed statistically significant increases in rejection, but valid-problem solving fell in three. Evaluations therefore need to score both versions and distinguish flaw recognition from the reported status.

Latent reasoning lets a large language model (LLM) think in a continuous space and verbalize only the answer. We argue that an effective latent thought must meet five requirements: it should be useful, helping produce the correct answer rather than merely changing it, diverse, so that resampling yields different reasoning trajectories, explainable, so that a decoded chain of thought (CoT) reflects reasoning the answer actually follows, refinable with more inference compute, and efficient, costing less than an explicit CoT at comparable accuracy. Current methods rarely meet these requirements: they learn shortcuts from the question, distill the explicit CoT into their weights, or imitate it one token at a time. We focus on flow matching in a learned latent space, the family we argue is best placed to meet them, and identify the training choices that make it work. The result is Flow-based Latent Reasoning (FLaRe), a simple recipe covering what the latent space encodes and how to shape it, where to train the flow, how to read out the answer, and a final stage of training on the model's own verified thoughts. A probe for each requirement shows that FLaRe improves on prior latent methods in all five. It also compares favorably with them on arithmetic benchmarks, while reaching 97% of the accuracy of explicit CoT at a quarter of its latency.

Workplace sensing studies combine long-running behaviour traces with self-reports, yet the tools that collect those data often sit apart from the interface that returns results. We present TrustmeWatcher, the application built for the TRUST-ME project to connect this work. TrustmeWatcher reuses ActivityWatch's OS-level watchers for computer-activity collection and adds its own application layer. It turns the collected traces into an interactive screen-time dashboard, synchronizes responses from short questionnaires completed on the StreamDeck, and presents questionnaires alongside video highlights. Activity records and self-reports are aligned into labelled records for model development. The scope of this paper is limited to ActivityWatch data as model input. Artificial intelligence (AI) uses these activity records to predict six normalized state scores and an overall well-being score. The trained model runs locally, and the dashboard presents its predictions in semantic bands. Explainable artificial intelligence (XAI) helps users understand how recorded activity contributed to a prediction. Privacy Control lets users pause or resume the camera and eye tracker used by the study. We describe the workflow, its user-device and sensing-setup boundaries, and its use with records from 17 participants. The result is a deployed application and study workflow that integrates activity review, study data collection, privacy control, local prediction, and a participant-facing interface for XAI evaluation.

Hyper-connections widen the residual stream of a Transformer to $n$ parallel streams. Their manifold-constrained version (mHC) mixes the streams at each layer with a doubly stochastic matrix, which it computes by Sinkhorn normalization of exponentiated logits. We give a geometric theory of this design on the Birkhoff polytope. First, a doubly stochastic mixer splits the stream into a mean channel, on which mHC is exactly a residual network, and a difference channel, which each layer contracts by its second singular value $σ_2 \le 1 - n \min_{ij} H_{ij}$. Thus the extra width is a fading memory with a horizon of $1/(1-σ_2)$ layers, and among nonnegative mixers only the permutations do not collapse. Second, the Sinkhorn-logit map is a global chart, and its logit gradient is exactly the Fisher-Rao gradient. Thus logit gradient flow follows a squared Fisher-Rao metric, and the straight-through update is exactly entropic mirror descent. Third, under logit gradient flow the logarithm of each entry moves at a rate of at most $4n^3\|\nabla f\|_\infty \varepsilon$, where $\varepsilon$ is the distance to the nearest permutation. Thus gradient flow approaches and leaves the vertices only at rate $1/t$, but mirror descent moves at an exponential rate. Fourth, the local convergence factor of Sinkhorn is $σ_2^2$, so a fixed iteration budget limits the horizon. Experiments confirm the predicted rates.

Suppose a committee, expert panel, or other group is making judgments on some issues, where these may be not just yes/no-questions, such as whether a defendant is guilty, but also variables with many possible values, such as macroeconomic or meteorological variables or travel directions. Furthermore, there may be interconnections between different issues, as in the case of economic or climate variables. How can the group arrive at "intelligent" collective judgments, based on the group members' individual judgments? We investigate three challenges raised by this judgment-aggregation problem. First, reasonable methods of aggregation (such as defining the collective judgment for each issue as the average or median judgment) can produce inconsistent collective judgments. Secondly, many methods of aggregation are manipulable by strategic voting. Finally, not all methods of aggregation are conducive to tracking the truth on the issues in question. We prove new impossibility or possibility theorems on all three challenges, identifying what it takes to produce collective judgments in a consistent, non-manipulable, and truth-tracking manner and thereby to achieve collective intelligence through aggregation. Overall, the median method, though imperfect, performs reasonably well. We also note the relevance of our analysis for non-human group decisions.

Methods for generalization in model-based reinforcement learning typically assume that an agent cannot recover the latent context governing the environment dynamics from its own experience, and therefore supplies it externally. We formalize and test this assumption with \emph{predictive sufficiency}, which quantifies what access to the context adds to next-step prediction under the visitation distribution an agent induces, and separates that quantity into a history-recoverable part, a residual requiring the true context, and the deficit added by a finite model. We classify context-aware algorithms by the predictive risk their conditioning set can target and demonstrate across environments of increasing identification difficulty that the headroom does not follow the MDP class. The same task under different priors leaves predictive headroom in one setting and nothing distinguishable from zero in another, where the agent's behavior implicitly identifies the context and any benefit of such a mechanism cannot be attributed to missing information. Where headroom persists, the learned state exposes it only partially, and adding the true context still lowers the risk. Our contribution is a practical criterion for matching contextual mechanisms to the information available to them, estimated from the ordinary trained agent without a reference policy.

Wikidata is one of the largest open knowledge bases, yet answering a complex question over it still requires a SPARQL query that names the right entities and properties and chains their relations. Language models offer a natural-language alternative but answer largely from memory, which is least reliable for less prominent entities. We study agents that instead answer by exploring the graph, and argue that two obstacles limit them: the lack of training data recording how a solver explores, and interfaces that add large graph results directly to the model's context. We test three hypotheses: that the difficulty of graph search can be controlled through the structure of a question rather than only through obscure entities or wording; that much of the failure on long-horizon search comes from how retrieved evidence is managed rather than from the model itself; and that, in a suitable environment, open-weight models can match commercial closed ones. We construct multi-hop questions on a frozen Wikidata snapshot by replacing named entities with nested conditions, checking after each expansion that the target remains unique and that every new condition is necessary. We release 10,235 solving traces over single-entity and multi-hop questions, together with the recursive language model (RLM) harness that produced them, in which models batch graph calls, keep results in persistent Python state and interpret selected evidence through sub-calls. On 100 questions, the harness improves both models we ran under both interfaces compared with direct tool calling over the same functions: gpt-6-luna rises from 49 to 61 correct answers, doubling its multi-hop accuracy, and Qwen3.8-27B, an open-weight model served on a single GPU, from 60 to 74.

We introduce a post-training method for diffusion language models (DLMs) that minimizes Maximum Mean Discrepancy (MMD) between generated and reference distributions in the feature space of a frozen pretrained DLM. To estimate MMD, we retain contextual features at individual token positions, obtaining multiple observations per sequence from a single extractor pass. We optimize this objective using policy gradients for discrete models and direct differentiation through generated latents for continuous models. In both cases, computing the loss directly from these features enables efficient post-training without full sampling trajectories or jointly trained auxiliary models. Experiments show lower generative perplexity at comparable entropy on OpenWebText and better accuracy-computation trade-offs on GSM8K. On 16B DMax-LLaDA2.0 models with hybrid masked-uniform diffusion, we increase decoding parallelism with similar or higher accuracy on math and code benchmarks.

Reinforcement learning (RL) has greatly advanced the capabilities of large language models (LLMs), but its memory demands remain a barrier to broader adoption. We introduce LoGRA, an approach to RL post-training that reduces memory by retaining useful learning signals in low-rank gradient sketches. These compact representations support both model updates and efficient policy synchronization. To prevent overly large updates from disrupting learning, we complement gradient compression with predicted-KL step control, which estimates policy changes before applying each update and adjusts its magnitude accordingly. Across reasoning tasks, LoGRA reduces average training memory by up to 45.7\% without sacrificing performance. It also enables stable training of a 27B-parameter model for over 1,100 steps on a single eight-GPU node, where dense Adam runs out of memory, making previously memory-infeasible RL training practical. Code is available in the \href{https://github.com/skzhang1/labs-molt/tree/logra/examples/scripts/logra}{Molt library}.

Robot learning in simulation depends on the objects the simulator offers. Many tasks need objects with separate parts, joints that allow the required motion, and physical properties that remain valid under contact. Existing methods recover this structure anew for every image: generative models predict parts and joints that mostly fail to settle or move in simulation, and general-purpose agents need a long session of model calls for each photograph. AffordCraft builds such an asset from a single RGB image and a task instruction by retrieval instead of generation: it locates the object and the part to operate, selects a matching entry from a library of articulated assets, and fits it to the image while keeping its parts and joints intact. Without any box or mask marking the object, AffordCraft produces a physically valid asset for 1,703 of 2,000 photographs from 31 categories. Five generative methods pass on at most 45% of the same photographs and, at the median, need 10 to 78 times our GPU time per valid asset. On 50 cluttered images, 162 of 237 annotated objects pass the same physical test after automatic detection. Growing the library from 141 to 11,372 entries needs no change to the method and raises category coverage from 46% to 100% and the share of selections with the requested label from 18% to 51%. We also build manipulation tasks from the constructed assets, both with single objects and in composed scenes; policies trained on scripted demonstrations complete both kinds of tasks from initial states unseen in training.

Machine learning (ML) studies using clinical data often rely on preprocessing and filtering pipelines before model development. The filtering decisions made in these pipelines can alter the dataset's statistical structure and may artificially reduce or increase the complexity of the prediction task. We argue that filtering choices should be treated as part of the scientific method rather than as a routine preprocessing step. We further discuss the need for explainable and transparent preprocessing pipelines that allow researchers to understand why specific filtering choices are made and how these choices affect the resulting data distribution and model performance. All of the source code for this work is available on GitHub.

Despite the promise and potential of quantum computing, most of the current day developments of quantum computing have several limitations for relevant practical use. As a result, hybrid classical-quantum computing has emerged as a viable solution. This paper discusses the software aspects of hybrid classical-quantum computing solutions to understand and examine how hybrid computing solutions can be expected to work in practice. This paper examines existing software solutions (cloud applications, middleware frameworks, and API gateways) that can enable classical and quantum computing to work together. For practitioners the paper provides a curated list of algorithms, full stack libraries, simulators, and cloud service providers that are likely to be central to the development of any hybrid computing solutions. We conclude by identifying an agenda for future research on classical-quantum hybrid software engineering, and some near-term priorities for development of functional, robust hybrid software systems. The paper argues that due to the deep dependency of quantum execution on classical pre- and post-processing, hybrid computing may represent a near-permanent outcome of classical-quantum computing rather than a transitional workaround.

World models must predict how an environment evolves under sequences of actions, enabling agents to compare possible futures and reason about counterfactual actions before acting. Long-horizon prediction is commonly obtained by recursively applying a one-step transition model, but intermediate errors can compound over time. Multi-step dynamics models instead condition on a sequence of future actions and predict their consequences directly, but become harder to learn as horizon grows: the model must track interacting state changes across the trajectory, endpoint supervision provides weak credit assignment, and intermediate predictions can remain plausible while losing information needed for later states. We show that these challenges can be addressed by casting the internal evolution of a multi-step transition as structured reasoning over textual world states: reasoning over sparse state changes reduces the burden of state tracking, a predictive-gain objective rewards the learned state for improving over a matched predictor that conditions on raw history instead, and intermediate predictive rewards supervise each state along the trajectory. Because these intermediate states are explicit textual representations of the world, they provide semantically meaningful targets that can be inspected, scored, and corrected during training. Across ScienceWorld, Jericho, and CEO-Bench, our approach achieves the strongest average long-horizon performance against recursive and non-recursive baselines that condition directly on raw history, with gains increasing at longer horizons. In a controlled counterfactual study, our model is also the only one with statistically significant sensitivity to future actions.

Many machine learning applications involve sensitive data and therefore require training under differential privacy (DP). However, DP training often degrades model utility. In some cases, first pre-training the model on "public" data before finetuning with DP on the sensitive data can reduce the drop in utility. However, the success of this depends on how relevant the selected public dataset is to the sensitive data. We introduce the first pipeline that privately learns the mixture of several public datasets to pretrain on for a given sensitive downstream task. Our key insight is that we can privately find the best mixture of multiple public datasets by privately learning a low-dimensional linear model. We tested our method on the NIH dataset for X-ray classification and the ENRON email dataset for language modeling. Applying our method to find tailored mixtures of X-ray datasets to pretrain on for diseases in the NIH ChestX-ray14 dataset, we improved macro AUC by up to 0.037 across privacy budgets compared to the baselines, with gains as large as +22.8% relative AUC on Cardiomegaly at $ε=1$. For DP training on the ENRON dataset, pre-training on our mixture of The Common Pile (a collection of public-domain text datasets) decreased test perplexity by 16% relative to the baseline mixtures.

CoVariance Neural Networks and their extensions have emerged as effective tools for processing multivariate data, deriving graph shift operators directly from second-order statistics. These architectures, however, are designed for independent and identically distributed observations and do not fully capture the joint structure of temporal and cross-variable dependencies in multivariate time series. In this work, we introduce Inverse Cross-Spectral Neural Networks (iCSNNs), a class of graph neural networks for stationary multivariate time series whose shift operators are the inverse cross-spectral density (iCSD) matrices. These operators encode frequency-specific conditional relationships among variables, exploiting the decomposition provided by the spectral representation theorem. Leveraging spectral smoothness, frequencies are grouped into bands sharing a single iCSD operator, yielding a compact parametrisation that retains the frequency-dependent structure of the process. We further propose a joint learning procedure to estimate both the Fourier-domain dependence structure and the iCSNN parameters, adapting the iCSD operators to the downstream task. When tested on synthetic data, iCSNN outperforms baselines from different methodological families.

The information bottleneck (IB) seeks a representation $U$ of a source $X$ that retains as much information as possible about a target $Y$, subject to a constraint on $I(U;X)$. A classical argument shows that it suffices to consider representations with at most $|\mathcal{X}|+1$ symbols, and this bound is known to be tight whenever $|\mathcal{X}| \geq 3$. We show that the binary case behaves differently: if $X$ is binary and $Y$ is finite, then for every joint distribution of $(X,Y)$ and every rate constraint, the IB optimum is attained by a binary $U$. Hence the bound $|\mathcal{U}| \leq |\mathcal{X}|+1$ sharpens to $|\mathcal{U}| \leq |\mathcal{X}|$ for binary sources. The proof combines a separating hyperplane argument with the observation that, for a binary source, the ratio of the second derivatives of the two entropy functions involved is concave.

Large language models (LLMs) annotate and scale political text or constructs by generating text tokens. A new class of models, which TypeSafe markets as "System One" models, instead returns decisions and probability distributions across a user-supplied fixed answer set. A commercial model, JEV, is advertised as having a dramatic cost and speed advantage over traditional LLMs along with better calibrated decisions. As such, it might be useful for social scientists looking to quickly and cost-effectively annotate or scale large corpora of text and have a reliable indicator of a classifier's uncertainty. Yet, the accuracy of these claims and the broader model accuracy in social science text-based tasks are not yet established. In this paper, we do just that and hope to establish the suitability of JEV for social science tasks. We compare JEV with LLMs and human coders from published research, and with a current mid-tier commercial LLM (GPT-6 Luna) and an open-weight alternative (Qwen3.8-27B). We find that JEV matches, or comes close to, the capabilities of both LLMs in a variety of tasks. However, we find no cost advantage over GPT-6 Luna at OpenAI's batch prices. Further, we find that, when each question is asked once, JEV's probabilities are better calibrated than GPT-6 Luna's token probabilities, but not consistently better than Qwen3.8-27B's. We conclude that unless researchers have a need for speed, JEV's only obvious advantage is ease of parsing the underlying choice probabilities.

Building on our earlier program-evolution workflow guided by large language models (LLMs), we study weight-five bivariate bicycle (BB) and perturbed bivariate bicycle (PBB) codes. The resulting catalogue contains 1,142 distinct code proposals, including 1,081 nonbaseline proposals attributable to LLM-generated programs. Across the catalogue, we certify connected Calderbank--Shor--Steane (CSS) realizations [[96,4,10]], [[140,6,10]], and [[180,4,14]]. A post-search comparison certifies seven imported Lin--Pryadko archive constructions. For leading parameter triples also represented in that archive, we provide exact distance evidence, explicit bivariate presentations, and verified component reductions. A basis-independent connectivity analysis identifies 409 of the 1,142 catalogue entries as disconnected and shows that 73.1\% of the classes with exact distance certificates contain repeated connected components. Algebraic analysis organizes the connected CSS classes into order-3, order-7, and order-15 cyclotomic-kernel strata. The strongest exact connected PBB parameter point is [[216,4,10]], attained by two distinct component classes. Among the 936 distinct CSS proposals from the LLM-guided campaign with a recorded positive distance, 816 (87.18\%) are certified at $d\geq5$. For comparison, three random-search controls each sample 6,444 CSS code proposals uniformly without replacement, using the same per-lattice and encoded-dimension sample counts as the LLM-guided campaign. In these controls, 4,672--4,785 proposals (72.50--74.26\%) meet the same criterion. The LLM-guided campaign has the higher certified yield, while the random controls cover more connected classes. Together, these results extend LLM-guided discovery to a more constrained code family and provide a reproducible structural and exact-distance account of its strongest candidates.

Spectral variants of low-rank adaptation (LoRA) choose both a subspace and which factor to freeze. We separate these choices by freezing the input factor A or output factor B on the top or bottom singular directions of pretrained weights, with learning rates selected separately. At rank 2, the same-band advantage of freezing A is larger than either within-factor band difference on all four task-model pairs with complete comparisons. Freezing B also trails comparable-budget free LoRA by 8-18 percentage points on five pairs spanning a formatting task and OpenBookQA. The A-frozen advantage persists in a single-GPU-model replication and within individual MLP module groups, including controls with equal or greater trainable counts for B frozen, and when A is frozen on a random orthonormal basis. The factor contrast weakens with rank. On OpenBookQA / Qwen2.5-1.5B at rank 16, PEFT's MiCA implementation trails comparable-budget LoRA by 3.08 points under a shared training recipe transferred from the MiCA paper. A trained oracle output subspace largely removes the low-rank deficit; partial warm-up gains recur across three direction seeds. The factor-versus-band ordering is descriptive; an approximate multiplicity audit weakens several earlier significance claims. These results extend known factor asymmetry by showing how its magnitude depends on spectral placement, rank and training conditions.

World Action Models (WAMs) incorporate visual representations from video generation backbones to guide action prediction. Recent efficient WAMs adopt Mixture-of-Transformers (MoT) architectures and compute video representations once for reuse by the action expert. However, intra-expert iteration (\ie, multi-step action denoising) and inter-expert waiting (\ie, sequential execution of the video and action experts) still limit inference efficiency. To this end, we present RealtimeWAM, an extremely efficient WAM variant with one-step action generation and asynchronous inference, addressing these two bottlenecks. To reduce intra-expert iteration, we propose Teacher-Anchored Consistency Distillation (TACD) to address a local-global error gap: low local consistency error alone does not guarantee accurate final actions. TACD supplements local consistency with explicit supervision from the frozen teacher's multi-step rollout endpoint, enabling accurate one-step action generation. Additionally, we propose Cross-Expert Wavefront Pipelining (CEWP) to eliminate unnecessary expert-level waiting. It overlaps the two experts through block-wise sharing of the video KV cache, synchronizing only immediately before the corresponding action attention consumes it. Extensive experiments across diverse benchmarks (\eg, LIBERO, LIBERO-Plus and RoboTwin) and model variants (\eg, Fast-WAM and Faster-WAM) demonstrate the superiority of RealtimeWAM. Notably, RealtimeWAM maintains near-lossless performance (\ie, $<1\%$ drop) across these benchmarks while delivering significant end-to-end speedup (\eg, $\sim25\times$ on H100). Our code and checkpoints are available via this \href{https://github.com/ModelTC/LightX2V/tree/main/examples/realtimewam}{link}.

Language models can score psychiatric questionnaires from speech, but agreement with self-report may reflect the questionnaire rather than the person. We introduce psychometric representational similarity analysis, a framework for comparing the structure of speech-derived scores, self-report, item wording and theory, and implement it alongside person-level construct scoring in COMPASS 2.0. We show how similarly worded items induce covariance without psychological signal. In pre-registered discovery and confirmation analyses of clinical interviews from 275 participants, language-derived symptom geometry resembled wording more than self-report, with no structure beyond wording detected by the registered tests. Geometric agreement with self-report survived assigning participants someone else's answers, whereas person-paired scores captured distress more than specific symptoms. Complementary analyses examined counselling quality and wording structure across 34 instruments and the Research Domain Criteria (RDoC) framework. These findings distinguish agreement about psychological structure from evidence that language-derived assessments track individual people.

As generative models and AI agents propose chemical reactions at a scale beyond expert review, feasibility verifiers decide which proposals enter synthesis planning. But do their decisions agree with chemists across different kinds of candidates? We introduce FREA, a benchmark of 751 reactions labeled by expert chemists under an explicit feasibility criterion, drawn from retrosynthesis model proposals, zero-yield experimental records, edits by large language models (LLMs), and five negative candidate generation methods. Our evaluation finds that no verifier leads across all sources: LLMs given only the criterion are competitive with dedicated verifiers, while forward models perform best on retrosynthesis proposals but reject most feasible edits of recorded reactions at the evaluated operating points. Looking beyond aggregate scores, both forward models perform below chance when separating infeasible alternative disconnections from feasible generated candidates. To study whether negative supervision addresses these weaknesses, we also release a corpus of over 14 million recorded reactions and generated negative candidates. In matched training comparisons, adding a mixture of generated negatives to forward training raises mean AUROC across sources, but these gains do not extend to retrosynthesis proposals. Varying the generation method further shows that the largest gain on generated candidates coincides with worse proposal screening. These findings motivate evaluating verifiers against experts across sources and designing negatives for transfer to model proposals.

We study contextual bandits in which a surrogate is observed after the action but before the learner decides whether to acquire the primary outcome that defines action value and regret. The value of acquiring the primary outcome depends on both decision relevance (how much the current outcome matters for comparing policies) and the residual uncertainty after observing the surrogate. The Audited Surrogate Bandit (ASB) learns a contextual policy while allocating a budget of $B$ primary-outcome acquisitions over $T$ rounds. ASB sets a pre-surrogate acquisition level from current decision relevance and, after observing the surrogate, redistributes that level using an estimate of that residual uncertainty. For a finite class of $N$ policies over $K$ actions, ASB incurs $\widetilde O[\sqrt{KT\log N}\{1+\sqrt{T/B}\}]$ regret relative to the best policy in the class. In a two-action family where the surrogate does not reveal the better action, a learner that observes the surrogate before deciding whether to acquire can achieve bounded regret, whereas any learner that must decide before seeing the surrogate incurs $Ω(T/B)$ worst-case regret under the same budget. Synthetic experiments show that both acquisition factors matter: ASB has lower regret than variants using only decision relevance or only residual uncertainty. On a KuaiRec benchmark of user-video interactions, the regret gap relative to relevance-only acquisition widens and then narrows as the budget grows.

Transformers asked to multiply multi-digit numbers in a single forward pass often fail, and interpretability studies of pretrained language models find arithmetic solved by input-range heuristics rather than by an explicit carry. We train small Llama-style transformers from scratch on 4x4 multiplication without chain of thought and find that the input format is decisive: inserting a space token between digits raises exact-match accuracy from 1% to 89%. Output positions are learned in carry-chain order, with the middle digits, which have the longest-range dependencies, learned last. Inside the model, the separator token that predicts each digit (its prediction slot) encodes the carry-in as an angle on a ring in the residual stream; examples with more distinct carry values fill more of the ring. Activation patching between examples matched on the column sum shows that this state is causally used before the last layer: patching the prediction slot alone transfers the source carry in up to 84% of cases after block 4 for one middle column of our best model, while for other columns the carry is first assembled at the neighboring answer slot before reaching its own. Remaining errors are almost always off by one, consistent with a small error on the carry or on the circular digit code.

Text from multiple sources can become interleaved into a single sequence when attribution metadata is lost, such as overlapping speech transcripts, document reading flows, or concurrent agent streams. We formalize this challenge as Word-Level Text Unmixing: given an interleaved lexical stream and source count K, recover the original source sequences while preserving every word occurrence and its within-source order exactly. Directly generating separated texts with LLMs can omit, duplicate, or hallucinate words, violating this exact-reconstruction objective. We therefore propose Evidence-Preserving Ownership Routing (EPOR), which decouples source-ownership prediction from reconstruction. EPOR adapts a causal LLM to predict canonical ownership routes conditioned on the mixed stream and prior routing decisions. At inference, completion-safe constrained decoding is combined with deterministic indexed reconstruction, yielding structurally valid K-source partitions that preserve every observed occurrence exactly once. We also introduce UNMIXBENCH, covering controlled synthetic mixtures, timestamp-derived speech from AMI and ICSI, layout-derived document streams from ReadingBank, and simulated concurrent digital outputs. Across five evaluation tracks, a 4B EPOR model achieves the lowest mean minimum-permutation word error rate among finetuned baselines, reducing the five-track mean by 22.3% relative to compact source-array generation and remaining competitive with zero-shot frontier LLMs. These results show that when lexical evidence is fully observed, separating ownership inference from lexical regeneration provides a reliable alternative to direct generation.

Automatically extracted narrative networks -- graphs with entities as nodes and their relations as edges -- have proven useful for revealing central narrative structures through salient entities and their connections (Tangherlini et al. 2020; Labatut and Bost 2019). But a narrative is more than those central structures that everything else revolves around. This work is concerned with the everything else: brief sub-plots, small clusters of descriptions, or associations between minor characters that go under the radar at the macro-level. We present an approach to unearth such peripheral structures. They involve rare entities with limited textual presence, overshadowed by dominant entities and lost among each other in the long tail of many but rare entities (Baayen 2001). We leverage the known tendency of pointwise mutual information (PMI, Church and Hanks 1990) to inflate for rare events, turning its weakness into a strength by weighting edges with PMI to foreground peripheral entity configurations. Communities extracted from the resulting network are structural traces of underlying narrative elements. We demonstrate the approach on The Lord of the Rings. From measures of how concentrated or dispersed a community's activations are across the text, a typology emerges that reveals that peripheral structures form more than a single class: episodic passages, echoing long-distance connections, and recurring threads each surface as distinct configurations. The approach is conceptually simple and surfaces fine-grained narrative details that are lost in abundance, though its deliberate amplification of weak signals comes with inherent sensitivity -- best understood as a lens for exploration rather than a robust extraction pipeline.

Action-conditioned robot world models must respond precisely to robot trajectories while preserving realistic visual dynamics, yet learning both from heterogeneous robot videos remains challenging. Simulation offers structured motion supervision, but appearance differences hinder direct transfer, and inaccurate simulation predictions can misguide real-video generation. We present SimForcing, a simulation-guided framework that uses simulation both as a source of transferable motion knowledge and as a controllable reference for prediction. First, we transfer motion knowledge from a simulation teacher through latent-motion distillation, aligning temporal changes in latent space to internalize motion priors while mitigating the influence of appearance differences. Second, we introduce multi-block simulation conditioning with condition dropout to exploit predicted simulation trajectories without relying excessively on their accuracy. Our simulation-conditioning classifier-free guidance scheme unifies these two ideas by balancing predictions based on internalized motion knowledge with those additionally guided by simulation latents. The jointly trained student generates both simulation conditions and real-domain videos, requiring no additional world model at inference. On Bridge, SimForcing achieves the best PSNR, SSIM, LPIPS, and FVD among the compared methods without external embodied pretraining. Evaluation on InternData-A1 further supports its applicability across robot datasets. Moreover, using our trained world model to initialize a vision-language-action model improves LIBERO success, suggesting its utility for downstream policy learning. \url{https://github.com/Wang-Xiaodong1899/SimForcing}

LLM agents increasingly execute complex workflows involving multi-turn reasoning, tool use, and parallel agents. Efficient serving requires decisions that span two layers with complementary information: the agent harness understands workflow dependencies, context lifecycles, and execution objectives, whereas the inference engine observes request queues, KV-cache state, resource pressure, and execution capabilities. Existing interfaces do not systematically connect these views, limiting workflow-aware execution. HEAR, a bidirectional Harness--Engine Pairing protocol for agentic LLM serving. HEAR standardizes how the harness communicates workflow intent and execution requirements and how the engine returns runtime state, capabilities, and outcomes. By separating protocol semantics from optimization policies, HEAR supports diverse coordination strategies without changing workflow or model semantics. We instantiate HEAR for online cache-aware runtime coordination and workload-aware execution-mode selection for agent roles. Across four conversational and research-agent benchmarks under memory-constrained, concurrent serving, HEAR achieves a $1.61\times$ batch speedup and reduces median time-to-first-token by $2.23\times$ on SCBench. Mooncake shows that workflow intent and live engine state provide complementary benefits across load regimes. On BrowseComp-Plus and DeepResearchBench, workload-specific configurations yield $1.23\times$ and $2.45\times$ end-to-end speedups, respectively, without observed task-quality degradation. These results establish HEAR as a reusable coordination substrate for efficient agentic LLM serving.

How much of a conference accept/reject decision would change if the same paper were reviewed by a different set of reviewers? Running a second independent program committee is the gold standard for answering this, but it is prohibitively expensive: done only twice (NeurIPS 2014 and 2021). We build an observational estimator of this quantity from public review data alone, calibrate it twice, and apply it to nine years of ICLR (2017-2025; 36,113 papers, 134,912 reviews). The estimator decomposes scores with a Bayesian ordered-probit model into paper quality and reviewer noise, maps scores to decisions with a logistic model, and simulates two independent committees (posterior draws B=1,000; committee sizes k=2,3,4). Estimated disagreement rates are 23-30% at k=2 and 18-24% at k=4; 30-50% of accepted papers would be rejected. External calibration: at the NeurIPS 2021 reviewer-count caliber (k=3), the simulated 2021 disagreement rate is 23.3% [21.7%, 25.0%] vs. reported 23.0% (bias +0.3pp); accept precision and committee correlation agree within 5pp and 0.04. Internal calibration: on 18,740 papers with 4+ reviews, random model-free 2+2 reviewer splits agree with the k=2 simulation within 1pp in 2018 and 2021-2025. Longitudinally, we find no robust time trend in reviewer noise over 2017-2025. The high accepted-paper flip rates of 2020 and 2021 have distinct mechanisms: the 2020 four-point scale compressed scores (23.7% of papers had zero within-paper variance), and a counterfactual shows coarsening the scale raises disagreement by about 7pp; 2021 instead combined the lowest signal-to-noise ratio in the sample with the most threshold-crowded acceptances. For the LLM era, a 2023 breakpoint test on within-paper score variance finds no break, but the design has almost no power, and no post-2022 review text or confidence data exist, so no LLM attribution is attempted.

AI voice assistants often use Automatic Speech Recognition (ASR) with LLM-based reasoning, yet existing systems struggle with regional British accents, including Scottish, Irish, and Welsh accents, since most ASR models are trained predominantly on American English voice data. Consequently, errors can carry through to the LLM stage, corrupting tool-call arguments and producing wrong or missing responses, which is especially costly in finance. Deployable ASR must also meet tight latency and memory budgets, making an accent-robust model choice even harder. We introduce CavaBench, the first internally collected benchmark of spoken financial queries, and use it to evaluate a range of ASR models and their end-to-end ASR-LLM pipeline behaviour across self-reported British accents. We find that WER strongly predicts downstream tool-calling accuracy ($r = -0.93$) but can fail to reflect task-level performance, with accent-related failures varying substantially across models and acoustic conditions. These findings guide the design of more inclusive, reliable voice-based financial assistants.

The release decision for frontier AI systems increasingly relies on cyber capability benchmarks, yet public vulnerability benchmarks can expose agents to previously published advisories, exploits, and fixes, making it difficult to distinguish prior exposure from capability on unseen vulnerabilities. We evaluate open-weight and proprietary AI models on exploit generation, vulnerability repair and subsequent attacks in five nonpublic software environments, including vulnerabilities we privately disclosed while they remained unpatched. Researcher-developed and reviewed deterministic graders, not LLM judges, determine task scores. Comparisons with 209 disclosed vulnerabilities and cryptographic challenges reveal substantial variation across systems and vulnerability types. Repair scores exceed attack scores in two nonpublic environments and fall below them in three. Passing an initial security test is also insufficient: another exploit succeeds in 92 of 524 non-independent defender test intervals after the initial exploit is stopped. These results motivate vulnerability-specific attack-repair comparisons and subsequent resistance tests.

World-model controllers rely on action-conditioned dynamics for prediction and planning, yet real control systems often execute commands asynchronously due to communication delay, packet loss, reordering, and actuator buffering. We study how asynchronous execution changes the action semantics assumed within world-model controllers, rather than treating it only as an external control disturbance. Through controlled interventions, we identify two architecture-dependent failure modes: planning-based controllers such as TD-MPC2 suffer from a future-action timeline mismatch between imagined and executed action sequences, while recurrent world models such as DreamerV3 can attribute observed transitions to commands that were not actually applied. Our analysis shows that TD-MPC2 requires the correct future action sequence during latent dynamics rollout, whereas DreamerV3 requires timely attribution of each transition to the action that generated it. Based on these findings, we introduce two lightweight execution-consistent interfaces, Future-Sequence for TD-MPC2 and Applied-Action Feedback for DreamerV3, that correct these mismatches without modifying the pretrained world models. Experiments across delays, packet loss, reordering, multiple control domains, measured network traces, and a process-separated asynchronous stack consistently support both diagnoses and the corresponding architecture-specific corrections.

Spike-and-slab regression is a standard Bayesian formulation of variable selection: it returns a posterior distribution over which candidate effects are active rather than a single selected subset, so that every candidate effect carries an inclusion probability. Its cost grows exponentially with the number of candidate effects, so the posterior can be enumerated exactly only when the number of predictors is small. Beyond that reach, the posterior has to be approximated, typically by Markov chain Monte Carlo over the model space, which requires a fresh run for every dataset and, within a fixed budget of steps, may fail to converge. We present LinearPFN, a prior-data fitted transformer network that amortizes spike-and-slab inference for linear models with main effects and pairwise interactions. The network is pretrained once on synthetic datasets, drawn from an explicitly specified prior, and a single forward pass over a new dataset returns posterior inclusion probabilities, posterior-mean coefficients and posterior predictive distributions with no per-dataset fitting. The prior is conjugate by design, so that the posterior for each fixed set of active effects has a closed form, and wherever the exact posterior is still computable by enumeration we verify the network's outputs against it. On real predictor matrices from published social-science datasets, with outcomes drawn from the prior so that the true active set is known, LinearPFN attains a higher per-dataset selection AUC and a higher F1 under the median probability model rule than five classical baselines. The lead holds when the coefficients, the interactions or the noise depart from the prior. Code: https://github.com/schiekiera/LinearPFN. Trained model: https://huggingface.co/schiekiera/LinearPFN.

While NorMuon has achieved strong empirical performance in large-scale pretraining by enhancing Muon with row-wise adaptive scaling, its underlying adaptive mechanism remains poorly understood. In this work, we provide the first systematic analysis of NorMuon's adaptivity, revealing that it originates primarily from orthogonalization-induced geometry rather than genuine optimization-relevant information. Under exact orthogonalization, the adaptive scaling factors degenerate into a single global scalar for square and wide matrices, while for tall matrices their variation arises from the non-uniform distribution of row energy after orthogonalization. Under approximate orthogonalization, the orthogonality residual introduces additional variation into the scaling factors, leading to the \textit{Orthogonalization--Adaptivity Paradox}: more accurate orthogonalization weakens adaptivity. We further show that NorMuon's rigid row-wise scaling is geometrically misaligned with the one-sided orthogonal structure of tall matrices. Based on the analysis of these limitations, we propose two core design principles that a desirable adaptive mechanism for Muon should satisfy. First, adaptive scaling should be decoupled from orthogonalization, with the scaling factors computed directly from raw gradients. Second, adaptive scaling should be aligned with the shape-dependent orthogonal structure of the polar factor, using row-wise scaling for wide matrices and column-wise scaling for tall matrices. By incorporating several other design considerations, including sum-based second-moment estimates, bias correction, and adaptive clipping of scaling factors, we obtain the Decoupled Geometry-Aligned Muon (DGA-Muon) optimizer. We establish convergence guarantees for DGA-Muon and empirically validate both our theoretical characterization of NorMuon's scaling degeneration and the superiority of DGA-Muon over NorMuon.

Large language model (LLM)-based agents can exhibit deceptive behavior during task execution, including hiding failures, fabricating results, or falsely signaling task completion. Existing monitoring approaches mainly detect deception after it appears in observable actions or outputs. In this paper, we investigate whether deceptive behavior can be predicted from an agent's internal representations before it becomes externally visible. We frame deception monitoring as a trajectory-level representation analysis problem and align agent trajectories around key decision points. Using hidden states extracted before these points, we show that future honest and deceptive outcomes can be reliably distinguished, with predictive signals remaining detectable several model calls before the final decision. We further characterize the temporal evolution of these signals: deception-related representations are weak early in execution but become increasingly identifiable as trajectories progress, while transferable structure can emerge before the strongest decision-adjacent signals appear. Finally, we intervene on the identified honest-deceptive representation directions during inference and find that activation steering reduces downstream deceptive behavior, suggesting that these representations influence agent decisions. Our findings indicate that agent deception is an evolving internal process that can be detected and potentially mitigated before it is expressed externally.

Brain MRI interpretation is a longitudinal clinical reasoning problem: radiologists compare serial studies, integrate information across MRI sequences, localize findings within volumetric anatomy, and translate this evidence into report-grounded assessments. Existing medical VQA and 3D imaging benchmarks capture important parts of this workflow, but often evaluate brain MRI through isolated images, static volumes, or ungrounded report-style answers, thereby obscuring failures in the evidence chain that support clinical validity. We introduce BrainTRACE, a report-grounded benchmark for evaluating whether vision-language models can trace the evidence structure required for longitudinal brain MRI interpretation. BrainTRACE contains 7,273 scored VQA instances derived from 1,778 longitudinal patients, 7,299 MRI studies, and approximately 29k co-registered 3D MRI sequence volumes. The benchmark is organized by five levels of clinical reasoning, from acquisition recognition to case-level synthesis, and by evidence demands covering longitudinal comparison, report-grounded references, multi-sequence integration, and volumetric spatial evidence. BrainTRACE supports rendered inputs compatible with standard VLM interfaces, a 3D-evidence condition, and a decomposed case-reasoning track that audits six steps in a longitudinal evidence chain. Evaluation of 20 VLM configurations shows that current systems can identify isolated visual cues but rarely compose them into grounded longitudinal interpretations. We release the benchmark specification, evaluation lists, scoring implementation, scoring rubrics, and audit-record format to support reproducible progress in brain MRI VLM evaluation.

Large language model (LLM) agents are increasingly capable of acting in complex tool-use environments, yet they often fail to recognize when tasks are infeasible and no valid solution exists. Recent work has formalized this reliability gap as the problem of agentic abstention, and existing approaches typically optimize a model or agent harness against a fixed set of tasks, leading to limited generalization to unseen failure modes. We introduce HERA, a framework for harness-environment co-evolution for agentic abstention. HERA consists of (i) a pipeline to automatically construct verifiable pairs of feasible and infeasible tasks by applying controlled environment mutations that transform solvable tasks into cases requiring abstention, and (ii) a co-evolution procedure in which performance failures on previous tasks are used to drive harness adaptation and generate new execution environments and tasks geared towards previous weaknesses. On held-out evaluation tasks, an evolved harness from HERA improves abstention accuracy from 61.7% to 83.3% while improving feasible-task completion from 68.3% to 76.7%, achieving the highest abstention and feasible-task completion among the compared methods. The resulting best harness transfers across 19 other LLMs, improving abstention accuracy by 15.3 percentage points on average without any model-specific optimization, and enabling smaller models to match the performance of more powerful models at an estimated 85% lower cost.

Population prompts are widely used to generate synthetic survey responses, but they combine information supplied at inference with associations already encoded during pretraining. We introduce an alternative construction that maps declared aggregate preference anchors into group-indexed choice policies. For each population, the signs of six Global Preferences Survey (GPS) coordinates deterministically label a shared bank of paired synthetic responses, and Direct Preference Optimization fits a parameter-efficient adapter to those comparisons. We evaluate the adapters on candidate World Values Survey (WVS) items using prompts that omit country names and distinguish four questions: recovery of the imposed labels, transfer of the anchor signal to new text, coherence between the GPS anchors and human WVS responses, and agreement between adapter and human scores. The adapters recover the imposed pairwise labels. On a purposively selected sixteen-country development panel, adapter trust scores completely separate the two GPS-sign groups and have a rank correlation of (0.74) with continuous GPS trust scores. Human-GPS and adapter-human associations remain unresolved on the same panel, and results for the other preference dimensions are heterogeneous. These findings show that an anchored policy can retain a declared aggregate signal without thereby reproducing human response patterns. The contribution is therefore both an inspectable construction and an evaluation framework that separates anchor transfer from human criterion agreement.

Modern online platforms commonly rank ads and organic content separately before blending them into a feed displayed to the user, overlooking externalities: an item's click-through rate depends on its surrounding content, not only on its own position. Recent learning-based feed generation mechanisms model some of these cross-type interactions to globally optimize for the whole feed's welfare. However, these approaches either fix the ordering of organic content, or lack exact strategyproofness guarantees for bidders. To combat these shortfalls, we introduce the Maximal-in-Range Transformer (MIRT) mechanism class, which uses a transformer to generate a range of candidate feeds that jointly order ads and organic content, and selects the welfare-maximizing feed in the range. However, there is a tension: strategyproofness requires the generated range to be bid-independent, even though a candidate feed's welfare depends linearly on the bids. Our key technical contribution is a reinforcement learning approach that incorporates both candidate generation and bid-aware selection into training, enabling a bid-independent transformer to learn to generate high-welfare ranges by accounting for both individual feed quality and the collective quality of the range. Additionally, we bound the pseudo-dimension of the MIRT class under hard attention, showing that near-optimal expected welfare is learnable with sample complexity polynomial in the transformer size and only logarithmic in the range size. Empirically, MIRT outperforms the previous non-strategyproof state-of-the-art feed models while remaining exactly strategyproof. Our results show that transformer-based auctions can deliver externality-aware whole-feed optimization without sacrificing exact incentive compatibility, removing a major obstacle to their practical deployment.

Human activity recognition (HAR) relies on transforming sensor signals into informative representations for classification. Although deep learning and handcrafted features are widely used, the role of representation itself is often not systematically isolated. Signature transforms provide a mathematically grounded way to encode temporal order and cross-channel interactions, but their value for HAR under a fully reproducible and leakage-aware framework remains unclear. To evaluate whether signature-based feature learning improves HAR performance compared with raw-signal baselines, and to assess the effects of embedding strategy, truncation depth, model choice, and sensor configuration. Experiments were conducted on the UCI HAR dataset using a fully reproducible pipeline with the original train--test split preserved and subject-disjoint validation to prevent leakage. Three representations were compared: raw flattened signals, time-augmented paths, and lead--lag transformed paths. Signature features were computed at multiple truncation depths and evaluated using multilayer perceptron (MLP) and Random Forest (RF) classifiers under identical preprocessing and validation procedures. A prior K-means-based feature reduction study was also reproduced for comparison. Signature-based representations improved performance when paired with RF models, the best configuration was time-augmented six-channel signatures at depth 6 using entropy-based RF achieving 0.858 accuracy and 0.859 macro F1, outperforming the strongest raw baseline (0.816 accuracy). Lead--lag representations were competitive at moderate depths but did not surpass the best time-augmented models. MLP models did not exceed raw baselines. Signature-based feature learning can improve HAR, but its benefit depends on alignment between representation design and classifier choice.

Clinical decision-support outputs can lack an au- ditable link between patient observations, encoded knowledge, conclusions, and recommendations. We present the CKG Clinical Explanation Engine, a downstream layer for a frozen, training-free clinical knowledge-graph reasoner that converts patient inference states and disease knowledge into typed facts, explicit rule-application traces, provenance-linked conclusions, and policy-licensed recommendations. The design separates measurement availability, representation completeness, and disease-specific activation; consequently, observed zero-activation evience is not treated as missing and partial representation is distinct from unobserved evidence. Optional language generation is restricted to symbolically licensed content. Across five usable workbooks (6,720 patients; 20,160 patient-disease traces; 1,021,440 feature-evidence rows), IG-range validity and knowledge provenance were 100%, numerical cross-sheet fidelity was 100% (120,960/120,960), and exported logical/report trace completeness was 100% (20,160/20,160). Availability representation consistency was 99.7028% (1,018,404/1,021,440); all 3,036 disagreements were confined to three systematic feature-cohort patterns. The corpus contained 86,783 observed zero-activation and 139,949 observed partially represented instances. A separate seeded 25-patient end-to-end audit completed without execution failure and passed all pre-specified trace, licensing, provenance, and state-consistency checks. These results establish structural and implementation auditability, not clinical correctness or utility.

We present Empirical Variational Autoencoder, a general generative framework for continuous-valued (i.e., non-vector-quantized) sequences. EVA is based on the evidence lower bound of the Variational Autoencoder (VAE) but learns autoregressive latent priors empirically from training data, which can be implemented only by an additional single linear layer on top of VAEs. By replacing the conventional standard-Gaussian constraint with the self-predicted priors, EVA significantly alleviates the latent distribution gap between prior and posterior which is typically observed in conventional VAEs, and leads to high-fidelity ancestral sampling for sequential data generation. Extensive experiments on image and sound synthesis demonstrate that EVA achieves competitive generation quality with autoregressive diffusion baselines despite its much faster inference time.

Low-Rank Adaptation (LoRA) has become a standard approach for parameter-efficient fine-tuning, yet a fundamental practical question remains unresolved: how should the adapter rank be chosen? An overly small rank may lead to a poorly conditioned optimization landscape, whereas an unnecessarily large rank sacrifices the efficiency that motivates LoRA in the first place. Existing theoretical analyses provide only limited guidance on this trade-off, and their guarantees are typically established under restrictive theoretical settings. We address this gap by developing a substantially sharper landscape theory for LoRA, building on modern results from nonconvex low-rank matrix sensing. Our central insight is that the appropriate adapter rank should depend on the quality of the data-induced optimization geometry, rather than on the model alone. To formalize this connection, we introduce LoRA-RIP, a data-dependent restricted-isometry metric that characterizes the conditioning of the cross-entropy (CE) objective along LoRA-relevant low-rank directions. We prove that sufficient rank over-parameterization, with the required rank explicitly determined by the LoRA-RIP constant, eliminates spurious local minima, thereby extending existing RIP-based guarantees beyond the classical 1/3 regime. This characterization further enables principled data selection under a fixed rank budget. Experiments across language and vision tasks support these theoretical predictions, showing that rank and data quality are two coupled resources that should be jointly considered for more efficient and reliable LoRA fine-tuning.

Spatio-temporal graph models typically encode each snapshot with a GNN and then connect the resulting representations through a temporal module. This space-then-time design is effective, yet it does not explicitly represent how a pattern moves across the graph. We show empirically that, for a propagating process, the same present field can lead to different futures when its recent rate of change differs, motivating an explicit representation of motion in the predictive state. We introduce WaveGSSM, a second-order graph state-space model that maintains two coupled latent states at each node, one for the current pattern and one for its temporal rate of change. A graph-wave transition updates the motion state through graph interactions and uses it to advance the pattern state, coupling spatial propagation and temporal evolution within a single rollout. We evaluate WaveGSSM on four temporal-graph benchmarks and global weather forecasting. It consistently achieves the best mean performance across the temporal-graph benchmarks and reduces the geopotential RMSE by 20.2% on average for 1- to 5-day weather forecasts relative to a backbone-matched snapshot model, while better preserving large-scale atmospheric patterns.

Magic-state distillation is a major resource cost in fault-tolerant quantum computing. The cost of a magic-state factory depends strongly on its failure rate, which grows with the number of input magic states. Although symmetry-restricted methods have recently made distance two searches tractable, distance three and above have remained elusive at moderate input counts. We develop symmetry- and AI-assisted methods to search this regime. We present a unified binary-matrix formulation encompassing both triorthogonal-code and direct circuit searches. We show that distance at least three is equivalent to nonzero, pairwise distinct syndromes, separating the choice of syndromes from the search for compatible output gates. We restrict the syndrome search using group symmetry and language-model agents, followed by deterministic solving and independent verification. Our searches yield 699 factory classes, including 564 new ones. These include factories for pure-T states and factories with entangled outputs comprising combinations of T, CS, and CCZ magic states. The pure-T factories [[63, 11, 3]] and [[850, 128, 6]] achieve the lowest overhead exponents we know among protocols with at most 100 and 1000 inputs, respectively, with $γ= 1.589$ and $γ= 1.057$. Our [[1715, 287, 6]] factory, with $γ= 0.998$, is the smallest known pure-T factory with $γ< 1$. We also provide a context directory of search briefs and campaign notes with which readers can train their own agents and tailor the search to their requirements. With these results, we begin constructing an active, open-source repository of magic-state distillation protocols for the quantum community, supplemented by our methods and data, for the practical fault-tolerant quantum computing regime.

A model under pushback can correct itself, capitulate, or hold, and one flip rate counts a correction and a capitulation alike. Using SycoLens, a modular replay protocol, we test how user pressure and evaluation settings shape measured flip rates. Each measurement is one stateless replay of an item, a committed answer, and one scripted user line in a fixed form. Every effect is read against a matched control with the line deleted. Pushback wording, committed text, answer format, boundary distance, and ground truth become factors of one instrument; earlier instruments vary one to three of them. Across eleven frontier models from three providers and about 760,000 controlled replays, which models look sycophantic depends on how the user pushes back. Lines that assert the opposite verdict and lines that challenge the answer without asserting one rank the models almost unrelatedly. Flip effects grow several-fold near a model's boundary, yet items answered identically in every screening draw still carry about half of the most-affected totals. On arithmetic tasks where the truth is known, one model re-derives and corrects itself under pressure while another abandons correct answers without written work. On the model tested, a planted derivation lowers release of the answer it argues for, true or wrong, where a bare stated value does not; the wrong answer is corrected much more often than the true one is abandoned. Under a yes/no readout the rankings come closer, entangled with a pressure-induced shift toward "no". One score per model therefore compares different behaviours across models and benchmarks. We condense these dependencies into a reporting profile; the instrument, records, and analyses will be released upon publication.

Neurons of the brain exhibit a rich repertoire of electrophysiology dynamics with the same repeated stimulus eliciting very different voltage responses from the same cell. One common approach in biophysically detailed models is to capture this variability through ensembles of deterministic parametrizations, at a cost of hundreds of thousands of CPU hours. Existing machine learning surrogates inherit the same limitation, where a stimulus is mapped to a single voltage response. We address this challenge by learning a conditional generative model for single-neuron electrophysiology, using flow matching with a velocity field conditioned on the input current. On biophysically detailed models of two human cortical interneuron types, the generated responses closely reproduce the electrophysiological feature distributions, spike-time structure, and excitability profiles, even matching the experimental recordings from the corresponding human cortical neurons. Near the firing threshold, firing and non-firing responses coexist at the same stimulus amplitude, and at high amplitudes, ensembles may split into low- and high-firing modes near depolarization block. We show that our model recovers both modes in each case, while a neural operator baseline suppresses spiking near threshold and blurs the gap between modes at depolarization block.

IoT encompasses diverse physical entities, from smart home devices to autonomous vehicles, creating a complex environment with heterogeneous security models. This heterogeneity makes IoT sub-systems vulnerable to various network attacks. Modern security systems must therefore be more robust to ensure security and privacy for IoT applications. A highly secure IoT system also demands real time insight, requiring data collection at the edge of the computing layer. This diversity calls for a unified security model applied at the foundational level. Edge intelligence offers a direct approach to handling device diversity. A key goal of edge intelligence in IoT is to extract insight from local data; security models can then use this data to build local node protections, and integrating AI models yields an advanced security solution. This research proposes a novel deep learning algorithm for effective intrusion detection at the edge, supported by a cloud-based IoT framework. We evaluate the proposed cross modal deep learning algorithm against baseline models. The contribution is a cross domain Deep Neural Network (DNN) algorithm for intrusion detection. The objective is to assess a multi-method deep learning model to detect intrusions in IoT systems at the edge via community detection with modeled attention. We evaluate GCT auto-encoder, a novel framework integrating edge intelligence to identify security flaws. The model significantly improves performance and efficiency. On a network intrusion IoT dataset covering multiple attack scenarios, it achieved 0.908 accuracy, reduced learning loss to 0.00156, and outperformed existing approaches.

While modern large language models (LLMs) have been trained to reason through verbalized chains-of-thought, the generation cost grows substantially due to suboptimal paths to reach the final answer. Furthermore, as new insights are discovered while observing various input queries (e.g. through self-reflection), limited mechanisms exist for carrying forward these findings to be applied to subsequent problems. One can view the list of such strategies or behaviors as a growing cheatsheet, with elements retrieved from this memory module at inference-time. In this work, we consider structured cheatsheets, with learned clusters of behaviors. We introduce a Hierarchical Dirichlet Process Gaussian Mixture Model (HDP-GMM) over behavior embeddings, which shares components across domains while allowing domain-specific mixing weights, and uses the posterior predictive to retrieve relevant behaviors for a query; we call this a $\textit{Bayesian Cheatsheet}$. This mechanism allows for cheap adaptation in an online test-time training (TTT) setting, softly updating the mixture's sufficient statistics following each sample and enabling the creation of new components when the synthesized behaviors are sufficiently novel. We demonstrate that Bayesian Cheatsheet achieves clear performance gains relative to existing memory modules across reasoning benchmarks such as AIME'25, Omni-MATH, and PhysReason, even in the cold-start setting. We show that the Bayesian Cheatsheet is an adaptively reorganizing memory module, as behaviors can be re-assigned to components through a single step of collapsed Gibbs sampling. Our findings highlight the value of Bayesian-inspired memory modules for effective test-time adaptation and the role of structure in metacognitive reasoning.

The growing adoption of large language model (LLM) agents creates a need for network administrators and security teams to audit agent behavior within organizational networks without inspecting private user content. Network traffic offers an observable source of evidence, but how much it reveals about agent tasks and operations remains unclear. Existing traffic datasets lack the joint task and stage annotations needed to evaluate this question. We introduce ANT (Agent Network Traffic), a dataset providing agent behavior information at risk, scenario, and behavior primitive granularities alongside network traffic. ANT contains 3,114 execution episodes across 20 tasks and five scenarios, comprising 276,417 bidirectional flows and 40,049 behavior primitive segments organized into 47 macro groups. We establish a benchmark for agent risk identification, scenario recognition, and behavior primitive classification using 13 representative traffic analysis baselines. The results show that existing methods recover useful but uneven behavioral signals. They struggle to identify risk when malicious workflows resemble benign tasks and to distinguish scenarios with similar traffic patterns. Primitive classification is more reliable for frequent macro groups and those with distinctive traffic patterns than for rare or semantically similar groups. ANT provides a common basis for developing more precise auditing and forensic analysis of agent behavior from network traffic. Our data and code are available at https://anonymous.4open.science/r/ant-main-suite-7BC0/.

Evaluating text generation requires measuring how well the generated distribution matches the data distribution. For autoregressive models, this is done by the perplexity. Diffusion and flow-based language models can only provide a likelihood bound, whose tightness differs between model families. Sample-based substitutes such as generative perplexity with entropy do not consider the distribution fit. We propose SOL, a distance between text distributions. Each sequence is represented by the empirical measure of its hidden states under a fixed transformer and the distributions of these measures are compared by the double sliced Wasserstein distance. We prove that SOL is a metric if the transformer is injective. Experiments show that SOL detects distributional failures, recovers expected model trends, and provides stable sample-based estimates. We put forward SOL to fill the gap in the current evaluation protocol used for non auto-regressive models. As a first step we use SOL to re-evaluate a variety of models trained on OpenWebText.

To evaluate the frontier, we must measure models not by what they say, but by what they can engineer and build in grounded physical environments. We introduce \textsc{ArtifactArena}, an open-ended platform where models face a physically grounded hardware-software co-design challenge: engineering fully functional robots to compete in a simulated arena. We evaluate a frontier model's zero-shot, verifier guided refinement, and open-ended physical design capabilities through three harnesses that refine their bot artifacts based on text descriptions, physics simulator feedback, and gameplay data. We benchmark these capabilities with an Elo ranking of frontier models derived from head-to-head tournaments between their artifacts. By releasing this framework and tournament infrastructure for ongoing community submissions, we establish a living, non-saturating testbed to continuously measure the expanding limits of open-ended intelligence in the physical world. Please visit \href{https://artifactarena.ai}{https://artifactarena.ai} for more information.

Deep learning has revolutionised weather forecasting in recent years, especially through atmospheric foundation models, which offer competitive skill for a fraction of the computational costs of classic physics-based models. However, most existing foundation models are deterministic, limiting the generation of large ensembles for accurate uncertainty quantification, extreme weather risk assessment, and long-range weather forecasting. Furthermore, these models incur a large, often prohibitive, computational overhead to train from scratch. To address these shortcomings, we turn a pretrained deterministic prior model, namely the Aurora foundation model, into a generative ensemble-prediction model. To that end, we introduce a novel generative method, Denoising Stochastic Interpolants, combined with a replay buffer for Stochastic Differential Equation (SDE) rollout, enabling probabilistic training of SDE trajectories. Our stochastic foundation model, Xaurora, is finetuned from the small Aurora version, yet it approaches the state-of-the-art on global ensemble metrics and is competitive with the large version of Aurora. Our method is parameter and sample efficient, and generates skilful 15-day forecasts in 13 minutes. Our results demonstrate that deterministic foundation models can be efficiently extended into even stronger stochastic models.

Proactive AI assistants continuously observe a user's activity and decide whether to provide new guidance or remain silent. They should provide appropriate guidance for the task, determine when to provide the next guidance based on task progress, and adjust the guidance level to the user's expertise and needs. Supporting these capabilities requires training and evaluation data that reflect procedural structure and capture how guidance should adapt to task progress and user needs. However, existing datasets either focus on detection-based proactive understanding or provide procedural guidance at a fixed granularity. Fixed-granularity guidance provides limited information about fine-grained progress and broader procedural context, making it difficult to determine completion and adapt guidance granularity. To address these limitations, we introduce the ProactiveCoach suite, comprising ProactiveCoach-Instruct for training, ProactiveCoachBench for evaluation, and fine-tuned VLMs with an adaptive guidance system. ProactiveCoach-Instruct provides hierarchically structured guidance at the phase, step, and action levels for learning task progress and procedural context. ProactiveCoachBench evaluates whether models provide appropriate guidance at the right time across different guidance levels and adapt when the requested level changes. We fine-tune pretrained VLMs on ProactiveCoach-Instruct and demonstrate its effectiveness across backbones. Compared with fixed-granularity supervision, hierarchical supervision improves overall performance across backbones by up to 9.6%p. We further build an adaptive guidance system by combining our fine-tuned model with a lightweight guidance router. Without additional fine-tuning, our system outperforms the in-context adaptation baseline by 57.1%p across four guidance-level transitions. Our project page is available at https://jinsuby.github.io/ProactiveCoach/.

Text-to-image generative models can produce highly realistic imagery but also raise concerns about harmful misuse. While safety mechanisms exist, systematic evaluations of their effectiveness against realistic attacks remain limited. We present a systematic evaluation of harmful content generation across five open text-to-image models using an automated pipeline that transforms legitimate news captions into unsafe prompts targeting sexually explicit content, violence/gore, harmful stereotypes, self-harm, and hate speech. We evaluate both standard models with built-in safety mechanisms and community fine-tuned variants that bypass content restrictions. A human evaluation of 1,500 generated images shows high harmful-content generation rates: 89.2% for gore-related prompts, 47.6% for sexually explicit content, 43.6% for harmful stereotypes, 46.0% for hate speech, and 34.5% for self-harm, predominantly through graphic violence. Models show substantial capability for generating violent and stereotypical content, while community fine-tuned variants are particularly vulnerable to sexually explicit prompts. Generation quality is largely preserved under harmful prompting, producing imagery of sufficient fidelity to pose risks for disinformation and abuse; FLUX.1-dev produces clearly realistic harmful images in 30.9% of cases. We further evaluate automated moderation systems and find substantial detection gaps that allow unsafe images to evade filtering. Finally, we assess synthetic image detectors and show that models trained only on benign datasets perform worse on explicit content, while more diverse training data improves detection, highlighting semantic distribution gaps in current approaches. These findings expose limitations in current generation safeguards, moderation systems, and synthetic image detection, highlighting the need for stronger defenses against misuse at scale.

Content-based image retrieval (CBIR) in neuroimaging enables the identification of structurally similar brain scans, supporting diagnosis, prognosis, and treatment planning; however, existing methods are often limited to small datasets, single brain regions, or coarse class labels, thereby restricting their clinical utility and generalizability. Here, we present NeuroCBIR, a framework for fast and flexible retrieval of both whole-brain and region-specific 3D T1w MRI scans. A total of 103 cortical and subcortical regions are extracted to enable both whole-brain and region-level queries. NeuroCBIR leverages latent representations learned by a variational autoencoder (VAE) combined with contrastive learning, producing scan-specific embeddings that capture anatomical patterns. These embeddings were evaluated for subject re-identification, zero-shot age prediction, and zero-shot multi-class pathology stratification. Re-identification performance was high across both whole-brain and brain-region levels (mean average precision across the top-5 retrieved images (mAP@5) >= 98.4%), with robust generalization across datasets and acquisition conditions. While NeuroCBIR is not trained for age prediction or pathology stratification, zero-shot evaluations for these two tasks demonstrate that the embeddings encode meaningful information for downstream tasks. Embedding extraction on a 4-core CPU required approximately 18.7 s per scan, whereas similarity search was effectively instantaneous (less than 0.01 s). NeuroCBIR is publicly available for brain MRI with more than 26,000 precomputed T1w MRI embeddings. It supports reproducible research, region-specific flexibility, and clinically meaningful personalized diagnostic support. The software is available at https://github.com/minnelab/NeuroCBIR.

When a large language model handles a multi-turn task and a user proposes a change but ultimately rejects it, the model should continue as if nothing changed. We find a surprising failure: merely mentioning a rejected change can derail task execution, even when the user's final intent remains unchanged. To systematically study language model behavior under evolving user intent, we introduce Intent-Eval, a controlled benchmark spanning tool actions, code, databases, and mathematics. Across diverse tasks, models are vulnerable to both rejected proposals and superseded requirements, consistent with mentioned-as-in-effect confusion: conversational content is treated as active requirements even after it has been rejected or replaced. Accuracy degradation can deepen or persist as interaction continues, highlighting the need to distinguish what has been mentioned from what remains in effect. Building on this insight, we propose Intent-OPSD, a decision-conditioned on-policy self-distillation framework with Teacher and Student initialized from the same model. The frozen Teacher provides active-intent supervision from the complete task matching the user's decision, training the Student on the full dialogue to follow active requirements reflecting user intent.

Multi-structure segmentation of the uterus is important for computer-assisted screening, diagnosis, and treatment planning of uterine diseases, where ultrasound and MRI provide complementary clinical information. However, developing a unified model across these modalities is challenging due to their substantially different image appearances, anatomical contexts, spatial resolutions, and label spaces. Moreover, existing datasets often define different segmentation targets, making joint learning challenging and potentially leading to negative transfer across heterogeneous tasks. To this end, we propose a Topology-informed Prompt-conditioned Universal Segmentation (TPUS) framework for segmenting multiple uterine structures across ultrasound and MRI. TPUS introduces a graph-based multi-dataset backbone comprising modality-specific stems and a modality-shared graph-based encoder-decoder to support modality-sensitive input adaptation, structural feature reasoning, and joint representation learning across heterogeneous uterine segmentation tasks. In addition, TPUS uses task-aware class prompts to condition the segmentation process for different datasets and label spaces, a dynamic convolutional adaptation module to generate task-specific output responses, and a topology-informed loss to encourage anatomically consistent predictions. Experiments on a uterine ultrasound dataset and a T2-weighted uterine myoma MRI dataset demonstrate that TPUS achieves Dice scores of 0.898 and 0.693 on the two held-out test sets, respectively, outperforming several generic and universal segmentation baselines. Source code can be accessed at https://github.com/YonghengSun1997/TPUS.

Multi-view learning jointly exploits multiple complementary representations of the same data and has become increasingly important in machine learning. However, the Python ecosystem lacks actively maintained, unified tooling for end-to-end multi-view workflows. In this paper, we present polyview, a Python package that provides tools for multi-view embedding, clustering, fusion, and view augmentation, as well as for handling incomplete views, all compatible with scikit-learn. The library offers a unified interface for composing heterogeneous multi-view workflows, including seamless transitions between multi-view and single-view stages. It is built around a core set of classes and utilities that enable composition of different methods and straightforward implementation of new ones. We illustrate the package on five real multi-view datasets and compare its components based on canonical correlation analysis with those of two established libraries. polyview aims to be both a practical toolkit for benchmarking and prototyping multi-view methods and a foundation for future research and development in this area.

Reinforcement learning (RL) has emerged as a promising approach for placement optimization, particularly when combined with graph neural networks (GNNs) that capture circuit connectivity. However, most learning-based placement approaches focus on floorplanning, macro placement, or global placement, while detailed placement refinement remains relatively unexplored. In this paper, we present GPlaceRL, an open-source graph reinforcement learning framework for detailed placement refinement. GPlaceRL represents legalized placements as graphs and provides a modular environment for studying graph encoders, policy architectures, reward formulations, and local placement actions. To demonstrate the capabilities of GPlaceRL, we conduct a systematic evaluation of proximal policy optimization (PPO) policies with graph attention network (GAT) encoders in a per-design optimization setting. Across five placement benchmarks, the best greedy evaluation results achieve HPWL improvements ranging from $3.27\%$ to $32.87\%$. The results highlight the importance of compact GAT architectures and flexible local action spaces for placement optimization. Overall, GPlaceRL provides a reproducible and extensible framework for systematic research on RL-based detailed placement refinement.

Graph neural networks (GNNs) are the standard tool for node representation learning and are increasingly used in high-stakes settings. Their message-passing backbone, however, can amplify topological bias, raising fairness concerns. We study group fairness at the level of downstream predictions for node classification, link prediction, and node regression, and bound the demographic parity gap for an arbitrary number of sensitive groups. Our node classification bound is provably no looser than the closest prior result. For link prediction, ours is the first bound on the parity gap of the deployed sigmoid-activated prediction rather than a pre-activation proxy, and for node regression we provide the first such bound. Across all three tasks, the analysis identifies two distinct sources of bias: the separation between group means and the within-group covariance of the final representations. Building on this insight, we embed fairness into propagation itself by augmenting the convex smoothing problem underlying APPNP with a convex group-mean constraint and a within-group covariance regularizer. Unfolding projected gradient descent on this problem yields FairProp, whose layers pair a propagation step with a closed-form projection and which provably converges linearly to the unique fair optimum. Experiments on three tasks show that FairProp, even with exact group-mean equalization alone, provides a strong inductive bias that achieves excellent fairness-utility trade-offs against strong baselines.

As AI agents increasingly tackle complex repository-level coding tasks, distributing work across multiple agents is a natural way to scale beyond the capabilities of a single agent. To coordinate their interdependent work, these agents share findings and agree on interfaces between modules. However, exchanged information often serves only as context, leaving individual agents to interpret it and incorporate it into subsequent work. Consequently, shared findings may go unused and deviations from interface agreements may go undetected, undermining the reliability and efficiency of collaboration. This motivates moving part of the coordination responsibility from individual agents to the execution harness. To make shared information actionable during execution, we introduce the Artifact-Exclusive Communication Protocol (AECP). AECP requires agents to communicate exclusively through structured artifacts and specifies how the harness processes them. The harness supplies findings when agents access relevant code, screens implementations for mismatches with recorded interface commitments, and requires affected agents to revisit revised agreements. These coordination steps become part of harness execution rather than actions that agents must initiate from prior messages. Across Doc2Repo, NL2Repo, and CodeProjectEval, using closed- and open-source models including Opus-4.8 and DeepSeek-V4-Flash, AECP improves average test pass rate by 28.2% and reduces average wall time by 16.5% relative to an agent team using free-form inter-agent messages. Artifact-exclusive communication also blocks the relay of malicious instructions between agents, reducing how often they reach other agents from 95% to 0% and how often those agents act on them from 40% to 0%.

KV caches are a major bottleneck in long-context inference and long-form generation with large language models. Existing training-free eviction policies largely rely on proxy importance signals, such as attention mass, to decide which past tokens to retain. We argue that cache compression should instead preserve the predictive behavior of the full-cache model, retaining entries whose removal would substantially change the model's output distribution. We propose Behavior-Preserving KV Cache Compression, a training-free framework that scores candidate evictions by estimating the compressed-cache logits induced by their removal and evaluating the resulting KL to the full-cache next-token distribution. Using pre-eviction forward statistics, the method avoids running separate masked forward passes for each candidate. Across diverse architectures and both prefill-time and generation-time compression, our method delivers substantial gains in downstream task quality over lightweight attention-based heuristics at matched retained-KV budgets, with the largest gains under aggressive compression. It achieves these gains with additional compression-time computation while retaining an end-to-end speedup over full-cache inference in our evaluated settings.

Closed-loop evaluation of end-to-end driving requires continuous rollouts that reveal how earlier decisions affect subsequent driving. However, existing benchmarks evaluate only short segments and fail to capture later consequences. Ambiguous directional commands also obscure the intended navigation objective. We introduce Odyssey, a closed-loop benchmark for long-horizon driving comprising 100 scenarios, each reconstructed from a 100-second nuPlan driving log to preserve the context of navigation maneuvers and traffic interactions. To provide a consistent navigation objective, Odyssey replaces directional commands with explicit standard-definition (SD) map routes that specify which roads to follow, while sensor-based planning determines local driving actions. Throughout these rollouts, diffusion-based refinement of 3DGS-rendered images reduces rendering artifacts along the ego trajectory. To assess how effectively planners follow these routes and prepare for upcoming maneuvers, we introduce SD Route Compliance and Pre-Lane Change Score. These assessments are complemented by RouteDS, which extends the Driving Score with penalties for SD-route deviations and failed lane preparation. We adapt state-of-the-art planners, including vision-language-action (VLA) models, and evaluate their navigation performance using these metrics. Odyssey highlights open questions in route representation and integration for E2E driving. Benchmark code and adapted baselines will be released publicly.

Modern graph generative models typically operate directly in the discrete graph space, explicitly generating node and edge variables, which can become costly as graphs grow. In this paper, we perform generation explicitly on latent representations of entire graphs obtained from a pretrained Variational Autoencoder with high reconstruction fidelity. The generated representations, obtained through flow matching, are then decoded only at the final step. Across molecular benchmarks of increasing size, our approach achieves strong validity and FCD while offering a favorable quality-efficiency trade-off compared with state-of-the-art explicit graph generative models. One of the main advantages of this formulation is that the graph representation only needs to be learned once, after which the same one can be reused across multiple generative objectives without retraining. We demonstrate generation guided by molecular properties and further introduce validity-aware generation though a classifier learned directly in latent space. All code will be made available upon acceptance.

As technology scales to smaller nodes, increasing current densities make electromigration (EM) one of the dominant reliability challenges in on-chip interconnects. Accurate transient stress analysis is needed to identify wires susceptible to EM degradation, but applying physics-based solvers across many interconnects remains computationally expensive. This paper proposes a physics-guided transformer framework for fast EM stress prediction in multi-segment interconnect lines. The framework converts each line into geometry- and DC-aware segment tokens and uses transformer attention to capture line-level context. A lightweight query decoder then predicts stress at selected locations and time instants. The model is trained with an objective that combines normalized supervised regression, linewise relative-$L_2$ loss, and physics-guided continuity and terminal-flux terms. Experiments on IBM power grid benchmarks show that the proposed model achieves relative-$L_2$ error below 8\% and reaches up to 2459.68$\times$ speedup compared with the matrix exponential~solver.

The rapid advancement of LLM agents has enabled systems to autonomously perform complex tasks through external tools, but their growing access to personal data introduces significant privacy risks. Existing benchmarks primarily evaluate LLM agent privacy through simulated trajectories and outcome-based metrics, limiting their ability to capture privacy risks arising during multi-step agent execution. In this work, we introduce AgentPrivArena, a framework for evaluating privacy risks in realistic LLM agent workflows. AgentPrivArena integrates authentic MCP tools and self-hosted services within a reproducible execution environment. We further propose trajectory-level privacy metrics that quantify unnecessary information access beyond final response leakage. Building on this framework, we introduce AgentPrivAudit, a runtime auditing approach for monitoring privacy violations during agent execution. Extensive experiments on state-of-the-art LLM agents reveal substantial privacy risks overlooked by existing evaluation paradigms, highlighting the importance of trajectory-level auditing for trustworthy agent deployment.

Time Series Foundation Models (TSFMs) achieve strong generalization by learning to reconstruct or forecast broad temporal patterns from large-scale time series during pre-training. Yet this strength can become a weakness for anomaly detection: TSFMs may model rare anomalous patterns as effectively as normal ones, allowing anomalies to be accurately reconstructed or forecasted and thus diminishing their reconstruction/forecasting error-based anomaly scores. This paper proposes $\underline{\textbf{N}}$$\textbf{ormality}$ $\underline{\textbf{C}}$$\textbf{onstraint}$ $\underline{\textbf{L}}$$\textbf{earning}$ ($\textbf{NCL}$), a lightweight plug-and-play framework that adapts pre-trained TSFMs for accurate anomaly detection without modifying their pre-trained parameters. Our key insight is to constrain the broad pattern space of TSFMs to the normal structure of a target time series, preventing their broad modeling capability from obscuring abnormal deviations. Specifically, NCL constructs a compact normality subspace from a few normal patch features and adaptively steers each patch feature toward normality within this subspace, guided by contrastive constraints that form compact and discriminative normality manifolds. The calibrated features are aggregated to reinforce normal components and fused with the original TSFM output, amplifying the discrepancy between normal and abnormal observations for the reconstruction/forecasting error-based anomaly scoring. Extensive experiments across diverse TSFM families and benchmarks show that NCL consistently improves anomaly detection performance, providing a generalizable framework for adapting TSFMs to anomaly detection.

VLM safety is commonly evaluated through input- and output-level classification. Such classification is necessary, but it does not reveal whether a safety state is accessible or controllable inside the model. We argue that multimodal safety evaluation should therefore report a \emph{controllability profile} alongside behavioral classification, separating representation-level detectability, cross-modal specificity, intervention sensitivity, and benign-preserving selectivity. Using implicit toxicity as a stress case, we instantiate this profile on LlavaGuard and Qwen3.5 with sparse feature decompositions. LlavaGuard admits localized handles with a narrow benign-preserving intervention range and modest downstream safety gains, whereas Qwen3.5 supports strong representation-level readout but no comparable selective-control regime under the tested operators. These results show that internal readout and controllability can diverge. Future multimodal safety benchmarks should therefore report not only behavioral safety metrics, but also whether safety-relevant internal signals can be intervention-tested and controlled within a validated operating range.

Foundation models (FMs) are increasingly being developed for general time series and physiological signals, yet their transferability to downstream physiological tasks remains poorly understood. This question is particularly challenging for electromyography (EMG), where signal distributions vary substantially across users, sensing configurations, acquisition hardware, and downstream tasks. We introduce EMG-FM-Bench, a systematic benchmark for studying foundation-model transfer and adaptation on EMG. EMG-FM-Bench unifies 20 public datasets with over 1 million EMG segments and evaluates nine pretrained foundation models across four questions: how pretrained models perform when frozen or fully fine-tuned, how much pretraining helps compared with training the same model from scratch, how well models generalize to new users with limited labeled data, and how performance changes across different EMG tasks. Across the benchmark, linear probing provides useful information about pretrained representations, but full fine-tuning can substantially change downstream EMG performance. Comparing each pretrained model with the same model trained from scratch shows that the benefit of pretraining varies substantially across models and is not universal. Performance decreases when models are evaluated on new users, while five-shot adaptation improves macro-F1 in 70.2% of evaluated model-dataset combinations but recovers only part of the lost performance. Model performance is highly consistent between upper- and lower-limb classification and remains strongly correlated with continuous EMG-to-text decoding. Together, these results provide a systematic view of when pretrained time-series models transfer effectively to EMG and how their performance depends on fine-tuning, user variation, and downstream task.

Deploying model predictive control (MPC) requires constructing or identifying a predictive model for each target system. Time-series foundation models (TSFMs) offer an attractive option thanks to strong zero-shot forecasting capabilities across systems. However, low forecast error does not guarantee that a TSFM captures the system's response to the alternative actions considered by the controller. We study this gap using residential heat-pump control as a test bed, measuring the agreement between predicted and ground-truth effects of control interventions. Importantly, we find that TSFMs can recover the system's input-response relationship when the context contains sufficient independent control excitation. Common fine-tuning pipelines and feature smoothing reduce, but do not eliminate, the need for in-context excitation. Our results indicate that current TSFMs used for predictive control require sufficiently informative control variation in the inference context. Initial closed-loop results show promise for shorter context windows.

Large language models (LLMs) trained to answer questions are natively poor at teaching. Reinforcement Learning (RL) against a simulated student is a promising approach to improve their pedagogy, but existing RL-trained tutors reward the student's success on the tutored problem with the tutor's words still in context. The reward is then easiest to raise by telling the student the answer, and a tuned penalty is needed to reduce telling. Drawing on learning sciences, we introduce a masked near-transfer post-test: the student is tested on an unseen variant of the tutored problem with the tutor's utterances masked, so the reward can rise only through what the student wrote in its own turns. This discourages cognitive offloading by the student and allows the continuous penalty to be replaced by two binary reward gates (factual correctness of tutor response, no solution handover). A leave-one-out ablation shows that the learning-gain reward on its own does not separate teaching from telling: the gates reduce solution handover while the near-transfer post-test improves out-of-domain transfer. Using these reward designs we develop Eduardo, a multi-turn RL recipe for training LLM tutors, and use it to train 4B, 9B, 14B and 27B models from two distinct LLM architectures. Our post-trained Eduardo-27B model matches Gemini-3.1-Pro on MathTutorBench and Claude Opus 4.8 on TutorMoments at 2.4-6.2x fewer thinking tokens than frontier models, which matters for interactive tutoring. Without being named in the reward, the model more than doubles its use of the push-for-justification teacher move while support fading (e.g., assigning independent work), whose payoff lies beyond a single-problem dialog episode, is trained out. We open-source our training environment, an 8,671-problem near-transfer dataset, and trained models for further development.

Around 30\% of global energy expenditure can be attributed to the building sector, where a large portion of energy-consumption could be avoided by repairing existing faults. Fault detection and diagnosis (FDD) software addresses this issue; however, its creation and operation also have an environmental impact. The magnitude of this impact is influenced by the diagnosis architecture, as different architectures and methods have different energy demands. Yet, simply considering the energy consumed by the software itself is not sufficient to assess its overall environmental impact, since the diagnostic performance, e.g., number of detected faults or number of faults missed, also contributes to its ecological footprint. In this paper, we propose an energy-consumption model that considers FDD performance and energy spend directly by the diagnosis software. In an initial experiment, we compare several FDD architecture families, i.e., rule-based, model-based, classical machine learning, and large-language-model-based, in simulation using performance and energy-consumption values collected from prior literature. The results show that considering the computational energy and accuracy of FDD can change the relative benefit of the different approaches. Computationally efficient machine learning methods, such as random forest, provide the largest net savings on smaller buildings, whereas more resource-intensive approaches, such as fine-tuned large language models, become advantageous as building size increases. Our findings suggest that overall energy efficiency depends not only on the computational demand of the FDD software, but also on its diagnostic performance and the scale of the building.

The high cost of functional molecular assays, and prevalence of missing modalities and unmatched samples in computational biology, create significant barriers to comprehensive multi-omic profiling, essential for capturing and reasoning over molecules, cells, tissues, and organisms. This work proposes a model that learns meaningful representations from multi-omics cancer data supporting the reconstruction of missing and unpaired modalities. Contrary to increasingly complex, larger models, e.g. Foundation Models (FMs), ARO prioritizes practical applicability in limited or incomplete data settings. ARO optimally reconstructs missing modalities (MSE of $0.15$ on the validation and test data in the Unmasked settings), with its learned latent embeddings enabling a downstream cancer classification task. Our findings indicate that analyzing diverse molecular layers as a single integrated system offers a reliable and cost-efficient approach, reducing dependence on large-scale experimental testing, while still supporting multi-omic exploration in limited data settings.

What happens when a legal AI model learns to look like a lawyer instead of reasoning like one? We fine tune Qwen3-8B with Group Relative Policy Optimisation (GRPO) against a proxy built from three surface features: citation count, legalese density, and response length. The model does not learn to reason more effectively. It learns to withhold commitment. Across 16 yes or no legal reasoning tasks from LegalBench (N=320), overall accuracy collapses from 0.500 (chance) to 0.072 (McNemar p < 10^-36), driven entirely by the rate of properly formatted answers falling from 0.900 to 0.109. The model stops committing to answers. Yet when it does commit, accuracy rises from 0.556 to 0.657, showing that the collapse is not a failure of capability but a strategic response: the model has learned that verbose responses packed with citations but empty of a direct answer score higher than terse correct ones. We term this the Saul Goodman effect, a policy that becomes maximally lawyerly while becoming maximally noncommittal, and prove formally that it is the optimal response to any surface feature proxy that attaches no penalty to abstention. We further show that 89.3% of citations produced after training are structurally implausible hallucinations, many of them subtly corrupted names of real landmark cases, constructed in effect to survive a casual read and fail under scrutiny. To detect this failure mode before deployment, we introduce three diagnostic tools: the Confidence Theater Score (CTS), the Citation Plausibility Rate (CPR), and the Regret Gap (RG). In a domain where a confidently wrong answer can constitute malpractice, the broader lesson is direct: a reward function that measures how legal a response looks will produce a model that is maximally photogenic and minimally useful.

We introduce Heuristic-Guided Fourier Fine-Tuning (HeuFouFT), a task-guided framework for selecting trainable frequency coordinates in Fourier fine-tuning. Existing uniform and Gaussian band-pass schemes allocate a limited spectral budget through fixed, task-agnostic rules. HeuFouFT instead searches for coordinates using downstream performance. A coarse intensity map from lightweight block-level probes initializes three metaheuristic optimizers: Genetic Algorithm with Simulated Annealing (GA-SA), Particle Swarm Optimization (PSO), and Cuckoo Search (CS). During search, a Random Forest filters each population so that only the top 30% of candidates proceed to proxy fine-tuning. On E2E with GPT-2-Medium, all three variants outperform random-uniform FourierFT, Gaussian band-pass FourierFT, and LoRA across five metrics. PSO further outperforms LoCA, the best-performing baseline, on four metrics while using 37.6% fewer trainable spectral coefficients. Once coordinates are selected, HeuFouFT requires only 15--18% FLOPs of Full FT. These results show that task-guided search allocates limited spectral capacity more effectively than fixed sampling. Our code is publicly available.

Predicting patient-specific survival functions is crucial for clinicians in making informed decisions about patient care and treatment strategies. Among the various models available, the Survival Forest has demonstrated significant effectiveness in numerous scenarios. In this work, we propose an ensemble method that leverages the strengths of the Survival Forest as the master model, complemented by several base models. This ensemble incorporates the Beran estimator, a type of kernel estimator, to enhance predictions of patient-specific survival curves. We evaluated the performance of our proposed model using four distinct healthcare datasets. The results highlight the superiority of our ensemble method over baseline models in both calibration and ranking across most datasets. The findings suggest that our approach offers a more accurate and reliable estimation of patient-specific survival functions, providing a valuable tool for clinical decision-making.

Numerous agentic workflows are based on a generator-verifier loop: a generator proposes candidates, a cheap verifier scores them, and the workflow terminates when a proposal is verified as good enough. The verifier typically proxies a more costly ground-truth oracle, and as the generator searches adaptively against it, false acceptances may accumulate. Proposals can pass the proxy but fail under the costlier ground-truth check. We study when to stop these loops while controlling the false discovery rate of the accepted proposals. Our construction introduces tools of independent interest in distribution-free statistical testing and conformal risk control, including analysis of $e$-values constructed through index betting and a novel conformal risk control procedure for non-monotone losses. We validate the approach in synthetic settings and on a protein-design benchmark.

Regional accent cues can be captured under matched conditions, but it remains unclear whether they persist between read and spontaneous speech. We study RVG1, with 500 German speakers from nine regions, comparing ten speech representations on regional classification and continuous geolocation under matched conditions and speaker-independent read--spontaneous transfer. Whisper performs best under matched conditions, reaching 0.489 nine-way UAR and 148 km median geolocation error, but drops to 0.11/0.18 UAR across transfer directions and 363 km geolocation error. Self-supervised models show a similar degradation, whereas speaker embeddings are less discriminative in-domain but more robust under transfer. This contrast is consistent across classification and geolocation. Across representations, robustness is associated with how little a representation shifts between styles (style-invariance), for which crossstyle speaker retrieval is an interpretable proxy. Age, sex, sentence-overlap, and duration controls do not account for the gap, although channel characteristics contribute. These results show that strong matched-condition performance does not indicate robust regional information.

A semantic decision engine such as Jev can return a valid answer and still miss a network deadline, select an infeasible action or leave the service unverified. We systematize 139 paper families by decision interface, execution path and check ownership. Fifty families claim that their engine fits a control loop or time budget, but only four support the claim with matched measurement. Across all 139, four report deadline attainment. The gap concentrates where the decision has no deterministic computation step. Those 72 families make 22 of the claims, none supported, and name a coverage owner in only two. Bounded tests under one event model show that each gap can reverse an admission verdict. A decision that meets a 10 s budget for every isolated request meets it for none once decisions queue ahead of replayed execution times. The same engine passes one coverage check and fails another. We derive a minimum reporting record, design rules and a research agenda for admitting decision engines to control loops.

Spontaneous movement analysis in preterm infants relies increasingly on markerless pose estimation (PE) to derive clinically relevant motion biomarkers directly from video recordings. Training accurate infant PE models requires large sets of manually annotated keypoints, and human annotation is inherently prone to error. Noisy keypoints (i.e., keypoints mislocalized with respect to their true anatomical position) are especially problematic in this clinical setting, since they can propagate as artificial artifacts into the reconstructed joint trajectories. Building on the small-loss hypothesis and training-dynamics-based sample selection established in the noisy-label learning literature, we propose a novel framework for detecting noisy keypoint annotations. A hybrid convolutional-attention model is trained to predict the anatomical category of each keypoint from its spatial coordinates and local visual features; the resulting cross-entropy training dynamics are then used to derive per-keypoint descriptors, which are partitioned into clean and noisy subsets via unsupervised clustering. We validate the approach on NeoPose, a newly collected dataset of 65 hospitalized preterm infants, under two realistic noise scenarios (random positional perturbation and left-right swapping) across multiple noise levels. Results show that the proposed approach achieves an F1-score of up to 91.9% in noisy-keypoint detection. The framework further generalizes to the heterogeneous COCO benchmark, where filtering CE-detected noisy keypoints from the training set also yields measurable improvements (up to 7.4 AP points) in downstream pose estimation accuracy at moderate-to-high noise levels.

Federated Learning (FL) enables collaborative training of models across institutions without centralizing sensitive data, making it well-suited for privacy-concerned applications, such as medical imaging. To protect FL model updates during secure aggregation, additive masking is commonly employed. However, its underlying classical key establishment is only computationally secure. On the other hand, physics-based Information-Theoretically Secure (ITS) key exchange introduces practical constraints: finite key generation rates and time-limited storage severely limit throughput and sustained training of uncompressed models. In this work, we address this bottleneck by developing an FL framework that integrates frozen backbones, knowledge distillation, and quantization. These techniques reduce communication payload and, consequently, key material consumption. Moving beyond simulation, we benchmark this framework on a real physics-based key distribution testbed involving a chest X-ray classification application. Our results show that key usage can be reduced by $\sim$35$\times$ while maintaining predictive accuracy. This prevents buffer depletion and key expiration, enabling sustainable FL training under physical key generation constraints.

Training-free model merging aims to combine multiple fine-tuned models into a single model without further optimization on labeled data. Yet, in transformers, independently merging individual layers can affect a shared attention computation because the query-key and value-output operators depend on composed matrices, overlooking the functional structure. Moreover, when merging earlier components, downstream components receive different activations than they do in the original model, thus, the merged and original execution paths no longer match. In this paper, we introduce Sequential Local Operator Alignment, a training-free method that merges transformers along the execution path of the partially merged model. Our method uses calibration data to estimate the local behavior of each functional component, aligns operators sequentially under the intermediate activation of the partially merged model, and subsequently factorizes the merged operators back into valid transformer parameters. We empirically show that this sequential step reduces error accumulation across layers. Furthermore, the proposed operator factorization step enables rank expansion, providing a principled mechanism for increasing multi-task capacity. We demonstrate that our approach generalizes across modalities, model scales, and varying numbers of tasks, from CLIP and RoBERTa to billion-parameter LLMs, and further extends naturally to the merging of LoRA-fine-tuned models. The results indicate improvements over strong merging baselines without requiring rank expansion, while optional expansion provides a further accuracy-inference-cost trade-off. Project link: https://akansh12.github.io/SLOA-Merge/

Thinking-enabled vision-language models (VLMs) report ever-higher accuracy on spatial benchmarks, yet final-answer scores cannot reveal whether a correct prediction reflects faithful spatial reasoning or a linguistic shortcut. We introduce SpatialChain, a dataset of 28,350 training and 899 test examples pairing spatially-oriented GQA questions with scene-graph-grounded reasoning chains, retained only when the generated answer matches the symbolic ground truth, and a two-axis evaluation combining objective chain-overlap metrics with a scene-graph-aware LLM judge that scores faithfulness and completeness independently of the final answer. Applied to nine thinking-enabled VLMs, the protocol surfaces three findings invisible to standard accuracy: (i) four of nine models achieve $\geq$79% VQA accuracy while exhibiting shortcut rates above 39%, i.e., correct answers whose reasoning the judge marks as unfaithful; (ii) chain quality significantly predicts answer correctness for seven of nine models, but the two exceptions (Claude Sonnet 4.6, InternVL3.5-8B) reveal qualitatively distinct failure modes, terse output vs. verbose-decorative reasoning, that benchmark accuracy alone conflates; (iii) SFT on SpatialChain improves Qwen3-VL-8B by +6.2 pp in-domain and reduces its shortcut rate to 22%, while a stylistic specialization effect on external benchmarks motivates replay-augmented training as mitigation. The faithfulness judge is validated against 198 human-annotated items, where judge-human agreement matches human-human agreement, and against a second judge from a different provider, which preserves the model ranking ($ρ$ = 0.88). Data, generation scripts, and evaluation code are released at https://github.com/spatialchain/SpatialChainBenchmark.

Agentic systems increasingly coordinate molecular-design tools, but it is unclear which layer of the stack limits outcomes on real projects. We developed MAGI, an open modular agent that authors objectives, launches and monitors optimization, interprets structure--activity relationships, and revises its strategy accordingly. MAGI generates molecules either directly through the LLM or by delegating to REINVENT 4, with scoring services interchangeable behind a common contract. We tested it across nine retrospective lead-optimization campaigns from three pharmaceutical companies, replayed under fixed temporal cutoffs. Both routes produced valid structures: LLM proposals stayed closer to local chemistry and reached comparable or higher primary activity in fewer operations, whereas REINVENT explored broader chemical space. Whether a campaign met its objective depended on the predictive models, not on the generation route: attainment followed model accuracy on the chemistry proposed, dropping once that chemistry moved outside the model's applicability domain. Separately, a blinded evaluation asked whether the MAGI's output could pass as expert work: chemists were not able to discriminate agentic proposals from held-out compounds, and judged the SAR reasoning broadly plausible yet incomplete. Together, these results position MAGI as a coordination layer pluggable into existing computational chemistry workflows. The ceiling on real projects, however, remains currently set by scorer applicability rather than by tool orchestration.

Machine-learned interatomic potentials extend atomistic simulations beyond the length- and timescales accessible to electronic-structure methods. However, the computational cost of equivariant architectures limits the local correlations they can represent in practice and therefore their achievable accuracy. Here we introduce FlashCart, which makes higher-order correlations affordable by combining generated GPU kernels with an architecture that recursively builds equivariant features and compresses them to a fixed width at each step. We express tensor products in independent Cartesian components and symbolically simplify them and their derivatives, producing fused kernels that often outperform optimized spherical counterparts. We then show that increasing correlation order improves accuracy more efficiently than increasing width, depth, or tensor rank. On SPICE-MACE-OFF, FlashCart models advance the measured accuracy-efficiency frontier: a model with $5.6$ million parameters achieves lower energy and force errors and $10\times$ faster inference than a transformer with $189$ million parameters.

As agents take on long-horizon tasks, users shift from making individual decisions to overseeing autonomous execution. Yet the volume of agent activity and the fragmentation of supporting evidence make it difficult to determine which decisions warrant user verification. We study monitors that identify consequential decisions and locate evidence to help users assess their implications. We introduce AgentMonBench, a software-engineering benchmark comprising three subsets that cover two complementary dimensions: alignment between requirements and behavior, and awareness of consequential autonomous decisions for verification. To support these judgments, we propose the Evidence-Grounded Behavior Graph (EBG), a training-free method that groups source-linked evidence into behaviors and organizes their relationships into a graph. EBG presents task-oriented views of this graph to help monitors interpret behavior in context. Experiments across eight models show that EBG improves decision identification and evidence localization in most settings compared with direct access to the original context. Further experiments show that EBG's evidence-localization gains persist across input scales and hyperparameter settings, while real-world applications illustrate its practical value for human oversight.

We consider the stability of multi-step reasoning processes, which have extensive applications in language models, including chain-of-thought and algorithmic reasoning. While longer sequences of reasoning can improve a model's generation capability at test time, the errors due to intermediate reasoning steps can accumulate in autoregressive generation, and thus grow substantially at the end. In this paper, we ask: What are the key factors determining the stability of multi-step reasoning? First, we show an inference error bound governed by the product of spectral norms of the Jacobians taken through the input space across generation steps. This product can be viewed as an error amplification factor, which could scale exponentially with the number of reasoning steps, serving as a quantitative measure of reasoning stability. Second, we analyze this measure in transformer models trained to predict simple tasks like linear and quadratic functions. We theoretically prove that the transformer model converges to a solution where the stability measure decays, thus yielding nearly zero inference loss over (arbitrarily) long steps. Finally, the stability analysis leads to several algorithmic implications for controlling the stability, through (i) chain-of-thought length compression that reduces the sensitivity of each step, and (ii) quantization-aware training that regularizes the input Jacobian norms. We validate the proposed algorithms by fine-tuning language models on graph-algorithmic reasoning tasks and symbolic state-tracking tasks. Across seven evaluations, our algorithms improve over baseline comparisons by 3.5% on average, and by 8.2% for longer-length inputs. Ablation analysis validates that the stability measure is drastically reduced by 3-8$\times$, confirming the regularization effect on the spectral norms of the (input space) Jacobians.

Tool-using language-model agents are vulnerable to indirect prompt injection because they must act on untrusted external content. Existing training-time defenses can reduce attack success rates, but often at the cost of general capabilities. We show that training-based defenses induce substantial drift in the model's output distribution, altering its behavior even in benign settings and providing a potential mechanism for utility degradation. We further identify a failure mode of these defenses: On benign tool-use tasks, the model refrains from a step needed to finish an authorized task, particularly when that step is indicated by a tool output. To address these limitations, we introduce RAISED (Robust Attack Invariance through Self-Distillation), a training framework that combines self-generation and self-distillation. The model first generates its own tool-use scenarios, with an emphasis on cases where task completion requires acting on legitimate guidance from tool outputs. Then, through self-distillation, the student is trained to match the teacher's clean-context behavior on both clean and injected variants of the same trajectory. RAISED substantially reduces the attack success rate of prompt injections in tool responses while, unlike prior training-based defenses, preserving utility on both agentic and general-purpose benchmarks.

Despite significant progress in visual tasks by Multimodal Large Language Models (MLLMs), geometric diagram understanding remains challenging due to the presence of sparse visual cues and ambiguous symbol-primitive associations. MLLMs may therefore rely on textual priors, producing interpretations that conflict with visual evidence. We introduce the training-free Criticality-Driven Visual Intervention Framework (CVIF), an inference-time method that localizes critical layers and executes visual interventions during the transition from evidence aggregation to semantic decoding. At these layers, a Geometry-Constrained Local Relation Reconstruction (GCLR) module selects and weights vertex-centered visual evidence, while an Adaptive Visual Steering Operator (AVSO) redistributes attention mass toward the selected tokens. Experiments on PGPS9K and PGDP5K show that CVIF raises Overall F1 from 77.85 to 85.58 and from 75.23 to 82.84, respectively, establishing a novel inference-time visual intervention paradigm.

We propose MOSEL (Multi-Objective Stackelberg Efficient Learning), a framework for a posteriori multi-objective optimization (MOO) in deep neural networks that recovers a full front of Pareto stationary solutions at the computational cost of standard single-objective training. MOSEL reformulates the problem as a bilevel optimization problem that leverages network modularity to decouple representation learning from objective-preference alignment. Casting the bilevel optimization problem as a Stackelberg game enables solving the original a posteriori MOO problem in a single forward-backward pass. As a result, MOSEL matches the time and memory efficiency of standard single-objective training while enabling scalable Pareto stationary front learning. Empirically, MOSEL uncovers diverse and optimal Pareto frontiers in strongly conflicting settings (e.g., fairness-accuracy). Remarkably, even in weakly conflicting regimes such as multi-task learning, it consistently converges to solutions closer to the utopia point, outperforming both standard single-objective training and specialized multi-task learning methods. These results highlight the broader potential of a posteriori MOO learning as a pathway to efficiently learn more diverse and robust representations, ultimately improving generalization.

Large language models (LLMs) have achieved substantial performance gains through increases in model size, training data, and computational resources. However, traditional scaling approaches produce diminishing returns, rising financial and environmental costs, and barriers to participation for researchers operating outside large industrial laboratories. This review examines the evolution of LLM scaling theory from empirical scaling laws to compute-optimal training, with particular emphasis on parameter efficiency, token utilization, data efficiency, and resource-constrained environments. Foundational work on scaling laws is synthesized alongside later research on compute-optimal training, data pruning, efficient architectures, quantization, low-rank adaptation, and edge-oriented optimization. The literature indicates a shift from scale maximization toward more deliberate allocation of parameters, tokens, compute, and hardware resources. At the same time, important empirical, theoretical, and methodological gaps remain regarding whether scaling principles established on enterprise-grade infrastructure generalize to smaller models and constrained computing environments. This review organizes these developments into a unified framework for resource-efficient LLM training and argues that future progress should evaluate efficiency not solely through model performance, but through the relationship among performance, parameter count, computational cost, token allocation, and hardware constraints.

Inference-time steering offers cheap, fine-grained control over a language model's outputs by estimating a concept's representation in activation space and shifting activations towards it. Existing methods build these representations from activation averages over contrastive datasets. These averages incorporate unrelated concepts and noise, and are dominated by a few tokens, meaning the activation transport encodes token-level rather than thematic concepts. In this work, we steer towards examples that most express a concept thematically, rather than towards an expectation over all. We identify these examples using influence functions, which estimate how much each data point contributes to a model's representation of a concept. Unlike simple model activation similarity, they incorporate the curvature of the model's loss landscape, allowing them to capture concept-relevant relationships beyond superficial token-level similarity. We then propose influence-weighted activation transport, which uses optimal transport to steer activations of non-concept text towards those of concept text, weighting concept examples by their influence scores. We evaluate on toxicity suppression (Jigsaw), object-based concept induction (OneSec) and truthfulness induction (TruthfulQA), outperforming existing activation-transport baselines. We track capability after steering using perplexity and MMLU accuracy, finding that our method improves steering while largely preserving model quality. We further show that influence functions capture concept-relevant information that activation-based methods miss with the two approaches ranking data points significantly differently. Together, these results demonstrate the value of curvature-aware influence information for activation steering.

Integrating environmental sensor data with leaf imagery is widely reported to boost crop disease classification accuracy. In this work, we reveal that these reported gains are often artifacts of dataset construction: because a single sensor reading is shared across many images collected in a single session (one farm on one date), multimodal networks can predict disease simply by memorizing session identities. Analyzing two widely used Korean datasets, the Crop Disease Diagnosis (CDD) benchmark and an AI Hub pest/disease dataset, we demonstrate that nearly all images share sensor values, with 91.9% of CDD test images having exact sensor duplicates in the training set. Remarkably, an image-free classifier given only timestamps matches or exceeds sensor-driven predictions across all seven evaluated crops, and matches the published macro-F1 of a state-of-the-art CDD fusion model. These results indicate that performance gains on standard random splits cannot be disentangled from session leakage. We propose that multimodal crop studies must evaluate on session-held-out splits and report performance against sensor-free date-time baselines to ensure genuine generalization.

Large Language Models (LLMs) are increasingly deployed for automated software vulnerability analysis. Binary classification alone is insufficient; practitioners need explanations to triage bugs and engineer patches. Standard practice relies on Chain-of-Thought (CoT) prompting, but free-form reasoning allows models to obscure logical leaps, hallucinated execution steps, and internal inconsistencies behind plausible prose. Our manual audit reveals that approximately 60% of correct vulnerability verdicts are accompanied by fabricated or unverifiable claims, and free-form explanations allow reasoning errors to evade LLM-as-a-judge evaluation. We present Vulnerability Explanation Reasoning Auditor (VERA), an automated framework for auditing LLM vulnerability reasoning. Rather than accepting free-form text, VERA asks models to output a Structured Reasoning Record (SRR) encoding tracked pointers, memory operations, and state transitions in machine-readable fields. A multi-stage judge audits each SRR against eight reasoning failure modes using deterministic checks, with LLM calls reserved for semantic interpretation. The standardized SRR schema also enables automated mutation testing to benchmark judges at scale without human annotation. Our evaluation shows reasoning flaws occur in correct verdicts just as frequently as incorrect ones, and VERA exposes 87% of reasoning errors that free-form LLM-as-judge systematically miss.

Sinkholes are a widespread geohazard in karst terrain. In Florida, soluble carbonate bedrock, shallow groundwater, and intense rainfall combine to make subsidence both common and spatially heterogeneous. Predicting where and when sinkholes will occur is difficult for two reasons. First, locations without reported sinkholes cannot be directly labeled or sampled as true negative locations. Second, the potential factors governing sinkhole risk span heterogeneous data modalities and therefore require careful integration within a unified modeling framework. We address both problems with our proposed model, a multimodal Cox proportional hazards framework for sinkhole susceptibility. Our contributions are threefold. First, we extend the proportional-hazards formulation to heterogeneous multimodal input through modality-specific encoders and a cross-modal fusion layer. Second, we treat unreported locations as right-censored rather than negative, avoiding hard-negative labeling and yielding continuous, time-aware susceptibility from the predicted survival function. Third, a statewide Florida case study with spatially blocked validation and ablation studies quantifies the benefit of multimodal integration. A Florida case study demonstrates that the proposed method effectively ranks sinkhole risk and produces a statewide susceptibility map that captures spatial variations in sinkhole occurrence.

Memory self-evolution uses task feedback to iteratively improve executable memory programs that store and retrieve information from past interactions. Existing approaches typically adopt holistic evolution, deriving revision directions from mixed feedback and judging progress by overall performance. This can obscure optimization directions and hide capability-specific gains offset by regressions elsewhere, leaving promising directions underexplored. We introduce capability-driven evolution, which extends search guidance from overall performance to individual capability dimensions, preserving promising revisions and expanding exploration beyond the boundaries of holistic evolution. We propose PrisMem, which uses dependency-aware capability selection to prioritize targets with potential cross-capability benefits and history-guided diagnosis to refine capability specialists. Trace-guided integration compares evaluated programs on paired differential cases, using their behavioral differences to consolidate complementary gains into a unified memory program. Experiments show that PrisMem outperforms the strongest baselines by 10.54 and 7.83 percentage points on BEAM-1M and LongMemEval-M, respectively, demonstrating its effectiveness on million-token histories.

Reinforcement learning post-training for language models relies on two reward designs: human preferences (RLHF, DPO) and binary verifiers (RLVR). Clinical question answering fits neither. Near-correct answers differ by a single substituted entity, and no executable check decides clinical correctness. We instantiate a soft verifier from a maintained controlled vocabulary: UMLS Concept Unique Identifier overlap (via scispaCy, set-level F1) gives a graded, externally specified reward computed without a model in the loop. We combine it inside GRPO with an entropy-normalised LLM judge, which covers the safety and evidence axes overlap cannot see, and a small consistency penalty on padding and repetition that keeps early-training samples scorable. This three-term composite improves over SFT on Phi-3-mini (3.8B) over MedQA by 2.9% on EM (0.700 vs 0.680) and 39% on Token-F1 (0.202 vs 0.145); on Llama-3.2-3B the corresponding gains are 14% on EM and 35% on Token-F1. We report Token-F1 as the primary metric because it credits partially-correct clinical content that EM discards at this open-generation scale. Main-table results are means over 3 seeds with standard deviations below 0.005. The method transfers to PubMedQA, where training on the PubMedQA train set with the same composite reward improves Token-F1 over SFT by 22% on Phi-3-mini and 17% on Llama-3.2-3B without retuning. A reward ablation on Phi-3, varying the judge-ontology split at a fixed consistency weight, attributes 3 EM points to the ontology term, the contribution that catches entity substitutions the judge cannot. Three negative findings constrain the design: DPO under random negatives underperforms SFT for strong-prior models but helps the weakest-prior one; PPO under a sparse neural reward diverges; GRPO with KL-in-loss collapses at 7B.

Forecasting suicide risk is difficult due to the high heterogeneity of patients and the low base rate of suicide-related events (SREs). We present Latent Similarity Gaussian Processes (LSGPs), which embed patients in a continuous latent space to jointly model similarity and forecast risk. By selectively drawing information from latent peers, LSGPs better capture individualized risk trajectories, generalizing nomothetic (pooled), idiographic (per-patient), and hierarchical frameworks. Our contributions are: (1) an identifiable two-channel Similarity Kernel; (2) proof that the standard model-fitting algorithm, mean-field variational inference, collapses LSGPs to nomothetic models, along with a fix; and (3) empirical results on intensive longitudinal suicide data showing LSGPs outperform nomothetic, idiographic, and hierarchical models for next-week risk forecasting, with the largest gains in forecasting first-occurrence SREs.

Large language models (LLMs) are increasingly explored for graph understanding and decision-making, while System One models such as Jev select directly from supplied options. However, the capabilities of System One models on graph-related tasks remain unclear. We introduce GraphDecide, a model-independent benchmark that combines structural task profiles, matched graph-text input contrasts and heuristic-proposal controls to diagnose graph decision performance. We evaluate Jev and related choice-based models alongside language-model baselines, covering fourteen model-interface configurations. Jev's results illustrate the benchmark's central distinctions: accurate adjacency recognition does not guarantee broader structural correctness, joint graph-text input does not consistently improve prediction, and feasible construction does not establish high solution quality. Its task contracts, candidate interfaces and scoring rules support comparison across native selectors and language-model adapters. Code and aggregate results are available at https://github.com/VictorYXL/JevGraphBench.

Isogeometric analysis (IGA) solves partial differential equations accurately on exact NURBS geometry, whereas neural solvers are mesh-free but often orders of magnitude less accurate and typically trained by non-convex optimization without error control. We propose IGA-KAN, which uses local Kolmogorov-Arnold networks, fitted in closed form, to improve the IGA solution instead of replacing it. An IGA Galerkin solve produces u_h; on every knot-vertex patch a Kolmogorov-Arnold ridge model is fitted to the strong form of the equation, the exact boundary data and u_h, and the models are blended by IGA hat functions. With fixed inner functions the fit is one batched linear least-squares problem, without optimizer, learning rate or initialization. An a posteriori safeguard, motivated by a maximum-principle bound, decides where local models are used, keeping the IGA solution elsewhere. On eight benchmarks with exact solutions, five from the literature and one also posed on a domain fitted to a brain slice from MRI, the method reduces the error of IGA, at an unchanged number of Galerkin unknowns, by factors of 4.2 to 90 in L^2 and 4.1 to 220 in H^1 on the reference meshes, and its L^2 error is 6 to 6x10^4 times smaller than that of the best Kolmogorov-Arnold network trained from scratch on the same equations with a fixed budget. In an inverse problem it recovers an unknown constant source from one noise-free observation 167 times more accurately than IGA. The gain is attributed to the superconvergence of local averages of the Galerkin solution.

Adapting general-purpose large language models to specific tasks requires substantial human effort in designing data and training strategies. Sustaining improvement is especially challenging because model updates change the error distribution, requiring strategies to be continually refined. We introduce ImproveAnyTask, an autonomous post-training harness that improves task performance under a limited compute budget. Drawing inspiration from gradient-based parameter optimization, the harness organizes adaptation into error attribution, update-direction selection, and executable model updates. It combines metric-level and case-level analysis to identify a focal problem, then investigates research-backed strategies and compares their reported gains and reproduction difficulty. The selected strategy is translated into training data and a training configuration, with small-scale execution checks preceding full post-training. Subsequent evaluation guides model selection and further adaptation, while validated strategies and scripts are retained for reuse. Across 11 tasks, ImproveAnyTask achieves mean gains of 18.29 and 11.97 percentage points on the Base and Instruct models, respectively, with a maximum gain of 41.96 points, under a 24-hour budget with resources equivalent to eight H20 GPUs.

In this paper, we evaluate open-source generative LLMs on legal Natural Language Inference (NLI). Legal inspectorial processes take place in specific domains and often deal with confidential data. This creates a need for working with local models that do not require labeled training data. We evaluate our models on the ContractNLI benchmark and two NLI4Wills datasets. We successfully reproduce the baseline for the task (Span NLI BERT) and we evaluate multiple open-source LLMs on the same task. We analyze the invalid rate of the models, and their stability across temperature settings and domains. Among the generative models, Gemma-4 26B performs the best, reaching an accuracy of 81.2%, even outperforming the supervised model on one metric. On accuracy, it is not possible to beat the supervised model with zero-shot approaches. Qwen-3.6 35B performs well on both ContractNLI and additional datasets in the legal wills domain. Our findings indicate that zero-shot, open-source, generative LLMs are a viable alternative for real-world legal NLI when no supervised data is available. Our code is available at https://github.com/fbaratov/contractnli-llms.

In deep graph neural networks, increasing depth enlarges the receptive field but often leads to over-smoothing, where node representations tend to align. We develop a unified, mode-wise stability framework for deep GNN propagation that provides a principled characterization of over-smoothing. By interpreting layer depth as time and layer updates as graph-coupled dynamics, over-smoothing can be understood as an undesirable dynamical synchronization of features, for which the master stability curve provides a theoretical tool to assess the stability of synchrony. Guided by this theory, we further propose Stability-Shaped Deep Graph Learning (SDGL) to mitigate over-smoothing in deep GNNs. SDGL has two complementary instantiations: one induces controlled Turing instability to replace synchronization with spatial pattern formation, and the other maintains stable near-critical propagation. Experiments on diverse node- and graph-level benchmarks demonstrate the improved depth scaling and consistent accuracy gains over strong baselines, including graphs exhibiting long-range dependencies.

While reinforcement learning with verifiable rewards provides reliable outcome supervision for VideoLLMs, sequence-level rewards offer limited token-level guidance. On-policy self-distillation addresses this limitation by conditioning a self-teacher on privileged information to provide dense token-level supervision. However, aggregating heterogeneous evidence within a single teacher context obscures cross-evidence agreement and conflict. A further challenge lies in determining whether teacher guidance should refine reward-based updates or provide corrective supervision for failed trajectories. To address these issues, we propose MeSD, a multi-evidence self-distillation framework for VideoLLMs. MeSD constructs three evidence-conditioned teachers with shared parameters, using the ground-truth answer as a common semantic context while separately incorporating temporal and spatial evidence. Given the same student-generated prefixes, MeSD evaluates evidence-specific preferences relative to the Answer Teacher and fuses teacher-common preferences with gated teacher-specific residuals. Furthermore, MeSD introduces Verification-Guided Optimization to classify trajectories as Success, Failure, or Indeterminate. For Success and Indeterminate trajectories, MeSD refines token-level advantage magnitudes while preserving reward-derived signs. For verified failure trajectories that contain the required evidence, MeSD applies failure-conditioned distillation, using reverse-KL correction toward the fused distribution. Experiments on multiple video benchmarks demonstrate consistent gains over reinforcement learning and self-distillation baselines.

Modern LLM training increasingly relies on heterogeneous data sources spanning different domains, tasks, preference distributions, and difficulty levels. We study dynamic minimax regret for group-distributionally robust LLM post-training under instantaneous mini-batch-only bandit feedback. The framework views the training as a two-player sampler-optimizer process: a sampler adaptively selects among data sources using bandit feedback, while an optimizer updates the model parameters using stochastic gradients from the selected source. We focus on the practically restrictive setting where source losses evolve with model training but historical data are not re-evaluated, requiring the sampler to track instantaneous worst-sources from stale partial feedback. We propose DUCB-OGD, a simple and scalable algorithm that couples a Discounted Upper-Confidence-Bound sampler with an Online Gradient Descent optimizer. The sampler maintains exponential moving average loss estimates and confidence radii based on discounted effective sample sizes, avoiding costly re-evaluation of past data or intrusive changes to standard training pipelines. For $K$ data sources and $T$ training steps, we prove that DUCB-OGD achieves a dynamic minimax regret of $\tilde{O}(K^{1/4}T^{3/4})$, which is optimal up to logarithmic factors for the undiscounted objective under our feedback model. Extensive experiments across supervised fine-tuning, preference optimization, and reinforcement learning show that DUCB-OGD integrates seamlessly into modern LLM training pipelines and improves worst-group robustness with negligible computational overhead compared with standard sampling baselines.

Vision-based autonomous navigation for low-cost robots remains a fundamental challenge, primarily due to the significant gap between simulated training environments and real-world operational conditions. Direct policy transfer from simulation is often ineffective, while training exclusively on real data is impractical. We propose a hybrid transfer learning framework that effectively bridges the sim-to-real gap by combining domain randomization with feature-level domain adaptation. Our method employs a dual convolutional variational autoencoder architecture with a shared decoder, trained on an extensive set of 45225 simulated images and a minimal set of only 4556 real-world samples. This architecture learns a compact, common latent representation space that aligns the distributions of both domains. The adaptation process is further enhanced by two complementary data augmentation techniques designed to expand the limited real-world data. Experimental evaluation demonstrates that our method achieves an average success rate of almost 91% on image classification tasks for real-world indoor navigation, significantly outperforming both simulation-only and real-world-only training. We validate these findings through a direct, real-world deployment, where the proposed policy successfully guides a low-cost robot in a reactive exploration task. Furthermore, we validate the model's efficiency through a rigorous computational estimation, confirming its suitability for resource-constrained embedded platforms such as the Raspberry Pi 4 and NVIDIA Jetson Nano. This work presents a practical solution for developing effective and efficient navigation policies for low-cost robotic systems.

Multi-billion-parameter LLMs now run on phones for inference, and training them on the device would personalize them without user data leaving the phone. Prior work has measured individual training steps of such models on phones, but not complete training runs, and not whether adapters trained on the device improve personalization. We present the first systematic characterization of a multi-billion-parameter LLM fine-tuned on a mobile device, covering memory, per-step time, thermal behavior, and energy. An iPhone 17 Pro can fine-tune a 3B-parameter LLM to a typical user within one battery charge, and the resulting adapters improve personalization as much as adapters trained on a server. Sustained training throttles the phone to about half its initial throughput, and none of the pausing or burst schedules we tested recovers it. Nearly all of each training step is spent in the frozen base model, most of it in the backward pass, which nine of the ten other runtimes we audited do not accelerate. Apple's MLX had a kernel for it that was never dispatched and was incorrect, and our repair, now merged upstream, trains an adapter 1.47x faster on a third less energy. On-device fine-tuning is feasible on current phones, and making it efficient requires runtimes and operating systems to treat training as a first-class workload.

CLIP-like vision-language models remain a cornerstone of multimodal systems, yet their scores stay near chance on directed spatial relations, such as whether one object is left of another. We call this failure readout blindness and analyze, theoretically and empirically, why deployed scores miss the direction: when scoring rules treat the subject and object symmetrically, direction cancels regardless of encoder training. Guided by this analysis, we introduce Antisymmetric Displacement Readout (ADR), which aligns caption words with image patches in the frozen features and scores each relation by the signed displacement between matched object centroids. Notably, ADR succeeds without additional training or learned parameters, thereby demonstrating that directional information remains in the frozen encoder. However, text and world priors can inflate accuracy, so we further introduce prior deflation, which measures the benefit of the image-text pairing as the grounded gain over a null that pairs each item with an unrelated image. Extensive experiments across encoder families show that ADR substantially improves over deployed scores, which remain near chance on most direction-balanced sets even for fine-tuned encoders. Compared with more complex readouts, ADR outperforms the evaluated MLLM likelihood readouts and is competitive with their chat inference at a small fraction of the computation. These results support our claim that directional information can be recovered from frozen features by an appropriate readout. Our implementation and evaluation kit will be publicly available.

Materials literature contains detailed experimental knowledge, but procedures, chemical entities and measurements remain difficult to aggregate because they are reported in heterogeneous forms and depend on process-specific context. We present SciKGExtract, a schema-guided framework that combines large-language-model extraction with chemical normalization and agent-based evaluation and refinement before knowledge-graph integration. We evaluate the framework on 176 atomic-layer-deposition papers describing zinc oxide (ZnO) and indium--gallium--zinc oxide (IGZO), together with an expert-annotated full-schema subset. PubChem normalization improves exact-match extraction F1 for every tested model. For ZnO, the best F1 increases from 0.591 for direct normalized extraction to 0.805 with agentic refinement, whereas the best IGZO result is 0.344, revealing the greater difficulty of multicomponent supercycle processes. Evaluation against a deeply nested schema containing 65 experimental properties and 155 quantitative measurement nodes further exposes errors in process segmentation and numerical assignment. These results show that chemical canonicalization and targeted agentic verification provide complementary controls for converting complex materials literature into reusable, machine-actionable experimental knowledge.

This paper presents CRAFTER, an automated framework for designing and deploying self-adaptive IoT systems using Causal Reinforcement Learning (CRL). As IoT devices increasingly populate pervasive computing spaces, smart environments are enabled with advanced monitoring and interactive services. The dynamic nature of these environments, such as fluctuating workloads and evolving application demands, poses significant challenges in maintaining consistent Quality of Service (QoS) levels of IoT applications. While existing self-adaptation techniques offer adaptive capabilities, they are often designed to deal with specific application domains, hindering the design of self-adaptive solutions that can be re-used across multiple IoT verticals. In addition, there is a lack of automated pipelines that act on identifying key performance drivers to take effective adaptation decisions. CRAFTER addresses these issues by using Causality as a formal framework for performance analysis of IoT systems. CRAFTER generates causal graphs to uncover dependencies among system components and guide adaptation decisions based on cause-effect relationships. Then, adaptation agents can leverage this knowledge to take more effective adaptation decisions in dynamic situations. Our experimental evaluation demonstrates how CRAFTER enables deriving causal graphs spanning diverse IoT use cases. Furthermore, we showcase how CRAFTER improves self-adaptation performance by 25% compared to state-of-the-art Reinforcement Learning-based approaches.

Dense affordance supervision is an appealing auxiliary signal for vision-language-action (VLA) policies, yet naively co-training an affordance head can severely damage instruction following. We present a controlled study of how to wire such a head into a modern VLA on the LIBERO benchmark. Our recipe reads the backbone through a stop-gradient and re-injects an intermediate head feature into the action expert via a learned bridge. The stop-gradient is a precondition: letting affordance gradients reach the backbone drops the policy below the headless base (85.5% vs. 93.1%). With the backbone protected, a same-budget 2*2 ablation over injection topology (concatenation vs. residual) and bridge initialization (zero vs. random) shows initialization is the dominant lever. The best wiring, an actively initialized residual bridge, reaches 96.2%, matching the far more elaborate three-expert AffordanceVLA (95.8%) with under 1% extra parameters. Two probes explain the mechanism: ground-truth affordances fed as an input hurt, and inference-time zeroing shows a lazy bridge acts only as a training-time regularizer while an active bridge becomes load-bearing.

Article 50 (2) of the EU Artificial Intelligence Act requires providers of generative systems to make synthetic outputs machine-readable and detectable, while qualifying the effectiveness, interoperability, robustness, and reliability by technical feasibility, cost, content-specific limits, and the state of the art. For free-form text, one important implementation route is the implementation of a generative watermarking procedure, which poses a compliance problem that is hard to address. Strong watermarking is impossible against adaptive removal, while ordinary edits attenuate statistical evidence, and unmarked human text may overlap distributionally with machine output. This article develops an auditable alternative. First, it defines a description-length robustness profile. A finite-sample bound shows that detectable bias decays and that the required sample size grows with the inverse square of the decay rate. This replaces an unidentified Shannon-entropy constant with collision entropy. Second, it constructs label-conditional conformal prediction sets with separate false-attribution and false-exclusion levels, reporting ``watermark supported,'' ``not supported,'' or ``inconclusive''. Coverage is obtained as a finite-sample result and is class-conditional under exchangeability. A small reproducible simulation of a tournament watermark confirms both claims and shows that the surviving-token rule overstates the tolerable edit rate roughly twofold. The resulting premarket certificate, signed detector report, and postmarket recalibration protocol operationalize the Commission's 2026 Code of Practice without claiming universal robustness.

The application of Reinforcement Learning (RL) in Electronic Design Automation (EDA), particularly for chip placement, has attracted considerable attention in recent years. While existing machine learning (ML)-based approaches have achieved notable progress, they predominantly focus on generating optimal layouts in a single attempt, often producing solutions that require subsequent refinement. To address this limitation, we propose RollPlace, a novel and generalized macro placement framework. RollPlace adopts a two-stage optimization strategy: generating initial placement solutions via machine learning methods or heuristic-based strategies, and refining these layouts efficiently by adjusting specific macros derived from the initial stage. This strategy circumvents the sequential generation constraints inherent in traditional RL-based placement methods. Furthermore, RollPlace seamlessly integrates Monte Carlo Tree Search (MCTS) to balance exploration and exploitation, and employs a rollout mechanism for efficient local search. Extensive experiments on the ISPD 2005 benchmark demonstrate that RollPlace outperforms state-of-the-art methods. Additionally, end-to-end experimental results based on OpenROAD across 19 benchmarks show that RollPlace excels in multiple metrics. The proposed framework offers a robust and scalable solution for addressing the growing complexity of modern chip design challenges.

Electroencephalography (EEG) based brain-computer interfaces enable direct brain-to-device communication for applications such as rehabilitation and communication. However, their practical utility is often limited as the non-stationary nature of the EEG data introduces distribution shifts across domains (e.g., sessions and subjects). Adapting machine learning models to be invariant to these shifts in an unsupervised way, without using costly labeled calibration data, would drastically improve the utility of EEG data. In this work, we use a classic generative model of EEG to study distribution shifts introduced by the domain-specific forward process, which is associated with factors such as head geometry. We theoretically show that such distribution shifts can be recovered solely through linear transformations on the Symmetric Positive Definite manifold. Building on this insight, we propose SPDAlign, an interpretable framework for promoting domain-invariant EEG learning. SPDAlign first aligns the domain-specific means and corrects global rotations across domains using a recent optimal transport technique called Wasserstein Procrustes. We systematically study the proposed approach through simulations and demonstrate its competitive performance on extensive public EEG datasets. Additionally, SPDAlign is a globally linear framework and is intrinsically interpretable, so that the framework can identify frequency ranges of interest, determine the spatial patterns reflecting source-sensor relationships, and address cross-subject variability.

We study deterministic and randomized midpoint discretizations of Langevin dynamics for a target $π\propto e^{-V}$, where $0 \prec αI\preceq\nabla^2V\preceqβI$ and $κ=β/α$. To achieve $\sqrtα\,W_2\leqslant\varepsilon$, we show that deterministic Heun uses at most $\widetilde O(κ^{4/3}d^{1/3}\varepsilon^{-2/3})$ gradient queries, and underdamped exponential midpoint uses $\widetilde O(κ^{5/4}d^{1/4}\varepsilon^{-1/2})$. The proofs exploit cancellation at stationarity and smoothing using techniques from Malliavin calculus, outperforming previous upper bounds based on standard couplings. At bounded condition number, a lower bound matches the $d$ and $\varepsilon$ powers of both deterministic methods. To contrast, for the randomized midpoint methods and Poisson midpoint with at least two grid points (both overdamped and underdamped variants), a simple Gaussian calculation yields a lower bound $d^{1/3}\varepsilon^{-1/3}$ to get an $\varepsilon$-close sample despite starting at a benign initialization. This shows surprisingly that in high dimensions, deterministic discretizations can outperform their random counterparts.

Humans can often acquire and synthesize complex, recursive concepts from minimal experience. Leveraging cognitive insights, we propose the Minimalist Machine, a framework for inductive program synthesis designed to model such conceptual learning. The system uses a compact relational subset of Prolog: Programs are searched within a fixed schema of body-free facts and two-body conjunctive Horn clauses. Recursion is not defined by a dedicated metarule. Instead, it emerges when a target predicate is reused inside the body of a learned clause. Inspired by a primary school curriculum, the model is taught through a human-curated, sequential introduction of new concepts in arithmetic. Starting from initially empty knowledge base, it first acquires simple structural predicates, then successor-based state transformations, and finally recursive programs for addition, subtraction, multiplication, and division. Ultimately, this approach yields the fully transparent, inductive reasoning trace necessary for human-like conceptual learning.

Sentiment analysis is a fundamental problem in Natural Language Processing (NLP). Standard sentiment classification for the Arabic language remains challenging due to the high volume of dialectal Arabic. To advance research in this area, this paper proposes the Shared Task on Sentiment Analysis and Swapping in Arabic Dialects (DialectSentEval), hosted with the Arabic Natural Language Processing Conference (ArabicNLP 2026). This shared task consists of two subtasks: Subtask 1 focuses on multi-class and multi-dialect sentiment analysis, requiring models to identify sentiment polarity across various Arabic dialects. Subtask 2 introduces a generative task for Arabic sentiment swap, challenging models to invert sentiment polarity while preserving core semantics. In this overview paper, we present the motivation, dataset creation, and summarize the main findings from participating models.

GAMBIT is an opening chess move in which a player sacrifices a piece, typically a pawn, to gain a positional advantage later in the game. Analogously, in multi-robot coordination, individual robots may need to forgo locally reward-maximising behaviours to improve overall team performance. Such self-sacrificial behaviours are difficult to capture with manually designed heuristics, particularly in dense, interaction-rich environments. Focusing on double-integrator continuous dynamics, this work studies how to learn such coordinated heuristics over motion primitives for multi-robot trajectory execution. Our framework, GAMBIT, first learns coordinated motion-primitive selection through imitation learning and subsequently fine-tunes the policy through reinforcement learning. We further introduce a safeguarded rollout mechanism with backup trajectories that guarantees collision-free execution at all times. Experiments demonstrate that GAMBIT substantially outperforms a range of baselines, including centralised motion planners and decentralised reactive planners, while exhibiting strong scalability. In particular, it coordinates over a thousand robots with planning latency below a few hundred milliseconds in continuous domains.

Wi-Fi channel state information (CSI) enables contactless presence detection and gesture recognition. Its high-dimensional complex-valued time series require input representations that preserve informative temporal variations during compression. We propose Trajectory-Guided Tokenization (TGT), which combines complex trajectory decomposition with asymmetric attention to construct compact continuous tokens. For each antenna link and subcarrier, an orthonormal Helmert transform decomposes short, ordered temporal patches into local-center and centered-trajectory coordinates. Keys are learned from the centered-trajectory coordinates, while values retain both components. Learnable queries aggregate subcarriers into frequency slots, which are fused into temporal tokens. Trained jointly from scratch, TGT with TokenMLP achieves the highest mean accuracy of 92.83% among all evaluated frontend-backend combinations on the self-collected dataset. Experiments on EHUNAM and Widar further support the applicability of TGT to cross-domain presence detection and gesture recognition.

Industrial content moderation must process massive message streams under tight latency constraints, yet most abusive language (AL) detection systems rely on sentence-level classification (ALC), which neither localizes abusive spans nor identifies who is targeted. We define Abusive Language Identification (ALI) as a sequence-labeling task that jointly extracts AL spans and target mentions, and assess whether this approach can be used for text moderation. On a pilot corpus drawn from a production moderation pipeline, we compare ALI with ALC on cross-domain generalization and implicit abuse, and we also evaluate AL and target span detection. ALI remains competitive with ALC while providing localized outputs for moderators, with a modest and configuration-sensitive advantage on implicit abuse. Exact AL boundaries and target spans remain difficult to recover. We complement this comparison with a qualitative analysis and discuss perspectives on complete target--span linking and on structured benchmarks for ALI.

Long-context large language models (LLMs) have demonstrated strong capabilities across a wide range of tasks, but the growing KV cache introduces substantial memory and inference overhead. Existing one-shot KV cache compression methods typically commit to irreversible eviction immediately after prefill, before any signal from actual generation becomes available. Our quantitative analysis shows that early queries from the actual generation stage provide attention signals that are more consistent with subsequent decode attention, with the largest single-step gain occurring at the prefill-decode boundary. Based on this observation, we propose DeferKV, which moves the eviction decision from the end of prefill to the first real decoding step and temporally combines prompt-side and decode-side observations, thereby better aligning KV importance estimation with subsequent generation requirements. DeferKV requires no additional training, draft model, or future-query prediction module, making it simple and easy to deploy. Experiments on LongBench, RULER, and Needle-in-a-Haystack demonstrate that DeferKV consistently improves model performance under KV cache compression while maintaining low inference latency.

Concept bottleneck models (CBMs) are neural classifiers that allow to explain their decisions via high-level concepts, potentially enabling understanding, steering and debugging. However, their explanations are often derived heuristically. Building on formal explainability, we argue they should also be faithful, i.e., not misreport which concepts actually matter. We show that, for widespread CBM architectures, including recent VLM-based variants, faithful explanations must include all concepts in the bottleneck, compromising interpretability when this is large. This result applies to both heuristic and faithful-by-construction formal explanations. To encourage the existence of compact faithful explanations, we suggest i) modeling concepts probabilistically as binary or categorical random variables (rather than logits), and ii) employing per-concept training-time sparsification via group lasso (rather than regular elastic net). We also extend algorithms from formal explainability to CBMs, and show they outperform natural heuristics in terms of guarantees and explanation size. Overall, our work warns against naive interpretability claims and provides formal conditions and practical strategies for ensuring CBMs are as interpretable as advertised.

We introduce a novel invariance for peptide tandem mass spectrometry data, unlocking self-supervised representation learning that improves de novo sequencing of peptides. This invariance exploits the physical relationship between precursor properties (mass and charge) and fragment-ion evidence, without requiring peptide sequence labels. We introduce dIon, which adapts the DINO framework with two latent prediction tasks, both recovering a clean teacher representation: one from a spectrum mixture, using the precursor as a selection query, and one from a partial spectrum with the precursor withheld. The first associates precursor information with fragment-ion evidence; the second prevents representational collapse onto that information alone. Mechanistic probes support both effects, and ablations show that the full objective performs best. Under identical end-to-end training, dIon initialization improves de novo peptide precision over training from scratch by 5.5 and 8.4 percentage points on the held-out MassIVE-KB and Kingdoms test sets, and by 2.3 and 4.8 percentage points with a larger supervised training corpus. The resulting models surpass fully supervised state-of-the-art de novo sequencing models on the diverse, multi-species Kingdoms corpus under the same greedy-decoding protocol. Without peptide labels, dIon learns strong native peptide-similarity geometry compared with other learned models; with limited peptide-supervised adaptation, it achieves the best retrieval and pair-discrimination performance across all representation benchmarks.

World action models (WAMs) have emerged as a promising paradigm for robotic manipulation. They act by first predicting how a task should be performed and then decoding the actions from that future. However, the remaining actions are invalid once execution drifts from the prediction. Simply replanning from the already out of distribution state rarely restores what the task still requires; existing execution monitors decide when to stop, but not what to restore. We observe that the answer is already in hand: the future the WAM predicted before acting depicts exactly the states it intended to pass through. We introduce FAVOR (Future Anchored Verification and Online Recovery), a lightweight framework that keeps these predicted frames as anchors and uses them for verification and recovery. An Anchor Verifier compares each observation with its anchor, together with the executed actions, to flag deviations that break the task. Anchor-Guided Recovery uses a vision-language model to turn the flagged anchor into a short corrective instruction. Under strengthened instruction guidance, the WAM executes this instruction to return to the intended future. It then resumes the task. FAVOR raises the task success of the base WAM from 97.85% to 98.10% on LIBERO and from 72.60% to 72.98% on LIBERO-Plus without modifying the policy.

Many agent runtimes give the agent a tool for editing its own configuration. Some of that configuration grants abilities, such as enabling a tool. Other parts set the agent's limits: which directories it may write to, who may send it messages, which network address it listens on, how callers authenticate, and the gate that blocks risky writes. If the agent can edit those limits, a single ordinary request can widen them. We study this in a deployed, model-agnostic runtime. We propose a rule: the agent may change fields that grant abilities, and may never change fields that set its limits. We enforce the rule as a containment floor inside the configuration tool and measure what happens with and without it. Without the floor, a frontier model wrote a protected value on 25 of 72 ordinary requests that gave it permission to change settings, often when the request never named the field. Prohibitions written in the system prompt failed in a predictable way. A prompt that listed the protected field names stopped every request that used those names (0 of 36 saved, against 17 of 36 with no prompt) and did not stop the requests that only described the goal (10 of 36 saved, against 8 of 36). A prompt that described the forbidden effects did the reverse. With the floor, 0 of 167 protected writes were saved, although the models attempted a protected write in 65 of those cases. A search for other routes through the tool found only one, a pinned shell, which the floor's scope statement already excludes. The study covers two models and a single agent. We state what that does and does not support.

Statistically valid estimation of racial and ethnic disparities often requires inferring the probability that an individual belongs to a particular racial or ethnic group given only their name and geographic location. The standard approach, Bayesian Improved Surname Geocoding (BISG), relies on group population frequencies for each name. Although the U.S. Census Bureau provides such information for common names and a limited set of racial categories, comparable data do not exist for many racial and ethnic groups and are rarely available outside the U.S. We propose the list-powered BISG ($\ell$BISG) method, which can be used to derive calibrated group probabilities from group-specific name lists. These lists may be compiled based on expert knowledge or generated synthetically using large language models (LLMs), and thus may be subject to unknown biases. Representing names as embeddings, we treat list membership as a proxy prediction task and apply a correction based on proximal inference to recover the target group probabilities. We validate the method on U.S. voter files with self-reported race, on the full-count 1900 U.S. Census, and on the Lebanese voter registry. We find that LLM-generated name lists yield accurate and well-calibrated probabilities as well as precise disparity estimates comparable to those obtained using methods that require name-race data. Thus, $\ell$BISG substantially broadens the applicability of probabilistic race and ethnicity prediction to settings where name-race data are unavailable.

Adapting vision-language-action (VLA) models to deployment-time distribution shifts is important for reliable robotic operation, but conventional first-order adaptation can exceed the memory budget of inference-oriented deployment platforms. Zeroth-order (ZO) optimization offers a forward-only alternative with inference-level memory, but accurate gradient estimation requires many perturbation queries, making naive ZO prohibitively slow for large VLA models. We present VLA-ZO, a framework for fast ZO adaptation that exploits the structure of VLA computation. By confining adaptation to the action side, VLA-ZO keeps the expensive vision-language prefix frozen and reuses its conditioning states across perturbation queries and optimizer steps, while schedule-aware prefetching hides state-transfer overhead. On LIBERO camera-viewpoint shifts, VLA-ZO reduces end-to-end adaptation time by 25.59$\times$ at $q=16$ and 32.54$\times$ at $q=64$ relative to baseline ZO, while improving average task success from 48.27% without adaptation to 58.17% and 63.58%, respectively. These results show that making ZO faster can make larger query budgets practical, providing a promising path toward resource-efficient VLA adaptation on deployment platforms.

Probabilistic Neurosymbolic Learning (PNL) combines neural predictions with symbolic reasoning, enabling end-to-end learning from final-output supervision without labels for intermediate concepts. A central challenge is probabilistic inference: state-of-the-art approaches often rely on materializing the logical provenance of a query, which can itself become a major computational bottleneck. We introduce Dynamic Probabilistic Neurosymbolic Learning (DPNL), an oracle-guided framework that avoids requiring complete provenance materialization before inference. DPNL lazily explores the space of intermediate assignments, while oracles resolve entire regions that can already be certified to produce or exclude the target output. We establish conditions ensuring soundness and termination. ApproxDPNL extends the same search with early termination while maintaining certified bounds on the exact output probability, providing controlled approximation guarantees. The oracle interface decouples inference from the representation of the symbolic component, enabling problem-specific reasoning within the same framework. Experiments on several neurosymbolic tasks show that DPNL and ApproxDPNL substantially extend the range of problem instances tractable by probabilistic neurosymbolic inference.

Open-ended scientific discovery often requires repeatedly proposing and evaluating candidate solutions. LLM-based systems can support this process by generating and refining executable solutions from verifier feedback. Methods such as TTT-Discover use test-time training (TTT) to update the solution-generating LLM from verifier feedback, adapting its generation policy to improve subsequent proposals on the target problem. However, this becomes expensive when reliable execution requires a large model, since training must maintain gradients, optimizer states, and policy statistics while repeatedly generating long, structured outputs. It also complicates credit assignment: outcome-level verifier feedback must jointly evaluate the high-level strategy and its low-level implementation. In this work, we introduce Guidance-TTT, which separates these roles. A compact guidance model is trained at test time to propose high-level strategic changes, while a frozen execution model implements them as complete executable solutions. At each step, the system selects a promising previously discovered solution, proposes a change, executes and verifies it, and updates only the guidance model using an adaptive group-relative RL objective. This concentrates test-time learning on short strategic decisions while retaining the implementation capability of a substantially stronger model without adapting it. Without web access, Guidance-TTT produces strong solutions across four distinct domains: combinatorial optimization (Polyomino Packing), heuristic programming (AHC058), machine learning (Lasso), and GPU kernel optimization (TriMul). Across these tasks, it outperforms the best solutions reported in prior work while remaining competitive with state-of-the-art results on public online leaderboards. Code is available at https://github.com/Human-Agent-Society/reef/tree/guidance-ttt-support.

Freeway on-ramp merges are major sources of congestion, causing significant economic and environmental costs. While Deep Reinforcement Learning (DRL) offers a promising solution for ramp metering, existing approaches rely primarily on aggregated macroscopic data. Connected vehicles (CVs) provide vehicle-level observations that can complement aggregate traffic measurements, but their limited penetration produces incomplete microscopic information. This paper proposes a hybrid observation representation combining macroscopic traffic measurements with a two-channel grid encoding observed CV presence and speed. A Dueling Double Deep Q-Network processes these inputs to select ramp-metering green durations. The controller is trained under varying traffic demands and CV penetration rates and evaluated against ALINEA and macroscopic-only DRL variants in SUMO. Across 50 matched evaluation scenarios, the hybrid controller under partial CV visibility reduces the reported total travel time by 11.4 % and mean spillback duration by 84.9 % relative to ALINEA. Evaluating the same trained policy with full CV visibility yields a further travel-time reduction of approximately 1.6 %. Analysis across penetration rates suggests that the performance gap decreases as microscopic observations become more complete. These results support the use of complementary macroscopic and sparse microscopic observations for learning-based ramp metering. The source code implementation of the model is available at: https://github.com/youcefMehamlia/Multimodal-DRL-RMC

Diffusion transformers (DiTs) achieve state-of-the-art image generation, but their sampling cost limits deployment. Quantizing both weights and activations to 4 bits reduces this cost, yet existing methods fall short in one of two ways. Calibration-based methods are tied to a specific checkpoint and prompt distribution, whereas data-free Hadamard rotation, effective for LLMs, loses quality on DiTs. We show that this loss has a structural cause. Adaptive layer-norm conditioning adds a per-token mean to the activations, and at the widths of the evaluated DiTs, the Hadamard rotations used by data-free methods cannot spread this mean uniformly across coordinates. A single dominant direction therefore survives the rotation and sets the quantization range. We introduce CentriQ, a calibration-free quantizer that centers each token before rotation and restores the mean exactly through a rank-1 full-precision branch, so that per-token scales follow in closed form without data. Weights are fitted under a robust $\ell_p$ objective that tracks the dense mode of each group and discounts heavy tails. Across three DiTs, CentriQ matches the quality of calibrated SVDQuant at 4 bits, whereas calibration-free weight quantizers with plain per-token activation quantization collapse or degrade substantially. CentriQ outperforms the strongest calibration-free method reported to date at 2-bit weights. It is also the first calibration-free method to retain usable image quality at 2-bit activations.

Partial differential equation (PDE) inverse problems are often ill-posed, making fine-scale details difficult to recover. We address this problem by introducing a learned observation-complementary latent representation that preserves reconstruction-relevant information and is combined with the observation to reconstruct the unknown field. Building on this representation, we propose OCL-PDE, a generative framework that encourages the observation to guide large-scale structure and the latent to supply complementary fine-scale details. OCL-PDE is built on a physics-aware autoencoder (AE) and conditional Flow Matching, supporting inverse reconstruction as well as forward PDE prediction. Experiments demonstrate improved reconstruction accuracy and fine-detail recovery compared with the evaluated baselines.

Foundation models can generate scientific code, but authoring a scientific simulator (an executable program encoding hypotheses about how mechanisms generate observable signals) requires iterative refinement. Scientific adequacy rarely admits a unique implementation or exact test, so simulators must instead be judged against limited real observations. We study this setting as scientific simulator authoring under weak empirical feedback, where distributional comparisons between simulated and real signals guide revision, and the target is the simulator itself rather than only its generated samples. We introduce SimAuthor, a persistent authoring harness that retains and revises executable simulators, separates scalar search scores from structured discrepancy feedback, and accumulates reusable implementation mechanisms. We evaluate SimAuthor on six biomedical tasks spanning cardiac and respiratory audio, photoplethysmography (PPG), and electrocardiography (ECG). Under a fixed 100-attempt budget, SimAuthor outperforms PUCT score search on all six tasks, generally outperforms textual-strategy optimization, and achieves the highest endpoint score on five of six. The authored simulators also improve on unseen recordings, transfer to independent pretrained representations, and yield substantial out-of-distribution gains in downstream ECG classification. Finally, 111 of 138 audited revisions alter program structure and account for 86.1% of the signed score improvement. These results suggest that persistent revision can progressively convert foundation-model knowledge into better executable scientific simulators from limited empirical evidence.

Large Language Models are increasingly used in software development processes that rely on defined states, rules, and approvals. This creates a tension between the flexibility of LLMs and the need for reliable and traceable process states. We introduce Controlled Non-Determinism, a process-oriented approach that allows variation without letting it directly determine the authoritative process state. We place LLM-based processing within a controlled variation space that is bounded by a Deterministic Envelope. Before execution, we define the process-relevant properties of the result and the exit conditions. We also record the task configuration. These elements remain fixed during execution. At the end of the variation space, an exit check determines whether the result satisfies the predefined conditions and may become the next authoritative state. Different runs may produce different admissible results while the transition to an authoritative state remains controlled. We illustrate the approach with regulated software development and derive design guidelines for LLM-based changes whose results can be checked before they become part of the authoritative software state.

The Lipschitz property of a deep neural network provides a direct measure of its sensitivity to input perturbations and, when explicitly controlled, offers a principled way to limit the propagation of errors and improve robustness. Over the past decade, Lipschitz-bounded layers have been incorporated into increasingly expressive and high-performing deep models, narrowing the gap between empirical robustness and formal, by-design guarantees of stability. This article introduces the fundamental concepts underlying Lipschitz-bounded neural networks, explaining the principles behind Lipschitz-constrained layers, the mechanisms used to enforce their bounds, and how they yield robustness certificates at the cost of a single forward pass. The tutorial concludes by discussing emerging and open directions, highlighting Lipschitz control as a general framework for offering guaranteed, by-design stability.

Language-model systems batch questions for throughput, but unrelated questions should not change a target's answer when its input and numerical execution are fixed. We study compression of the key and value cache, which stores attention representations reused during generation. With request-local groups, Transformers' Half-Quadratic Quantization (HQQ) backend updates compression parameters separately but uses a shared average error to decide when all updates stop. Replacing only the question batched with the target changes four-bit HQQ answers in 170/384 test comparisons across two models. Replaying the other execution's update counts reproduces its complete answer and cache fingerprints in every changed pair, in both directions. Computing the stopping mean in FP32 reduces cache differences but leaves answer changes. Native HQQ also changes confirmed numerical correctness in eight arithmetic pairs. Fixed iterations and request-local stopping remove observed companion dependence under matched controls. Request-local stopping remains sensitive to synthetic padding changes at the tensor level. Fixing the original iteration budget removes this decision path without tuning. Neither repair has an established quality advantage, and natural rebatching still changes answers. Request-independence audits must cover stopping decisions as well as quantization groups.

World models, which learn how environments respond to actions, are emerging as a powerful paradigm for planning through imagined futures, transforming decision-making across games, robotics and autonomous driving. Bringing this capability to manufacturing could enable process decisions on timescales inaccessible to high-fidelity simulation. Here we introduce a generative world model for localized highly dynamic laser melt pool that predicts evolution from histories of temperature and phase morphology under candidate actions. Its generative latent dynamics capture the effects of unresolved melt flow, enabling more accurate recursive rollouts than deterministic regressors under transient laser inputs. Because the learned dynamics are differentiable, the model can serve directly as a predictive control plant. Gradients through imagined futures optimize laser schedules that regulate melt-pool depth over previously unseen geometry, path, initialization. We further distil this optimization into an amortized policy that produces control actions in a single forward pass, providing a proof of concept for real deployment on machines.

Scientific language models often access literature through untyped text chunks, which fragment the functional and evidential structure required for mechanism-rich questions. We introduce an evidence-grounded mechanism knowledge substrate that organizes scientific literature into provenance-linked evidence units, role-typed entities, and directed mechanism paths. We instantiate it as MS$^3$, a Material-Sensor-Signal-System schema for conductive-fiber flexible sensors, over 13,689 papers, 131,083 evidence items, and 26,648 mechanism objects. On in-domain and coverage-shift question-answering benchmarks, we compare closed-book generation, Web search, Raw-PDF RAG, and MS$^3$ retrieval across ten language models. MS$^3$ improves macro-averaged scientific correctness. It also improves citation entailment and answer completeness. These results support mechanism substrates as a reliable representation layer for scientific language models and motivate a source-repair workflow in which insufficient MS$^3$ evidence triggers targeted retrieval from its linked papers rather than assuming that a user has already supplied the correct PDFs.

Planar graphs are central to applications across science and engineering, yet existing generators provide limited support for goal-directed generation under hard structural and geometric feasibility constraints. We propose a dataset-free method for generating planar graph embeddings by combining parametric graph grammars with safe reinforcement learning to optimize generic task-specific objectives while satisfying constraints during construction. We formulate the generation process as a constrained Markov decision process, where the graph grammar defines the state and action spaces. We further introduce an action projection that maps sampled actions toward state-dependent safe sets, improving constraint satisfaction during training. In contrast to classical graph generators and deep generative models, which typically offer limited goal-directed control or rely on weak constraint satisfaction, our method constructs feasible planar graph embeddings directly during generation. We also introduce a benchmark suite for constrained and goal-directed planar graph generation, together with classical and deep generative baselines. Across all benchmark tasks, our method consistently outperforms baselines while satisfying the formulated constraints.

Parameter-efficient fine-tuning (PEFT) reduces the cost of adapting foundation models by focusing training on a small parameter subset. Complementary to this idea, we introduce RoSA (Rotational Sparse Adaptation), which narrows adaptation to a subset of layers at a time. RoSA freezes lower layers close to the input throughout training and rotates a trainable block over later layers, progressively increasing the number of frozen layers close to the input. This design reduces optimizer-state memory, shortens backpropagation, and even forward propagation if activations at the last frozen layer are cached. Because RoSA is orthogonal to the choice of trainable parameterization, it can be combined with PEFT methods or sparse optimizers within each active block. Experiments across multiple LLM architectures and tasks show that RoSA reduces peak memory while maintaining strong fine-tuning performance.

Wearable and bedside electrocardiogram (ECG) monitors must adapt to patient-specific morphology to maintain arrhythmia detection accuracy across users, yet personalization is typically performed offline and cannot account for individual physiology, electrode placement, or recording drift. On-device adaptation by backpropagation is expensive for microcontroller-class medical devices because it requires an optimizer state, repeated backward passes through convolutional layers, and labeled arrhythmic beats that may not be available at deployment time. This letter proposes prototype-only head adaptation as a compact personalization primitive for TinyML ECG systems. A one-dimensional convolutional neural network (1-D CNN; 1,314 parameters and 72.6k multiply-accumulate operations per beat) is trained offline on the MIT-BIH Arrhythmia Database under an inter-patient protocol, frozen as a feature extractor, and exported to a PSoC 6 microcontroller. Patient-specific adaptation then reduces to computing closed-form class means in a 32-dimensional embedding space, requiring no convolutional backward pass, no iterative optimization, and only one forward pass per support beat. Prototype adaptation improves inter-patient macro-F1 from 0.635/0.639/0.646 to 0.731/0.771/0.797 at 1/5/10-shot, outperforming linear stochastic-gradient-descent (SGD) head fine-tuning at every shot count for the target tiny backbone. On-device replay over 18 one-shot episodes on a PSoC 6 Cortex-M4F matches the host macro-F1 for the prototype head (0.798), with 11.39 ms per beat, 5.2 KB flash, and 22.2 KB SRAM. A restricted variant that updates only the normal-class prototype from passively buffered sinus beats yields a consistent +0.05 macro-F1 gain, reducing the annotation burden during initial

Cognition and culture have long been treated as parallel objects of study, joined more by metaphor than by mechanism. We argue they are co-constituted in a manner now empirically tractable: subjectivities can be measured as local trajectories through a high-dimensional cultural field, and the field is itself the aggregate of those trajectories. Recent advances in machine learning, including embedding methods and large generative models, provide the first general framework for measuring this co-constitution from micro-cognition to macro-social structure. Vector geometry recovers individual conceptual structure, organizational communication, and large-scale ideologies; captures multimodal cultural content beyond text; and generates testable predictions about cultural emergence, including the simultaneous arrival of independent discoveries across distant minds. We treat this predictability of simultaneous innovation as evidence for the co- constitutive view: when the geometry of the cultural field is measurable, trajectories of cognitive search through it become predictable. We then examine generative AI agents as simulators of human subjectivity and as a qualitatively new coordinating substrate whose insertion into social life introduces evolutionary dynamics unprecedented in human cultural history. We formalize this reflexive condition as a Heisenberg-like Uncertainty Principle for generative AI: as instruments for fixing a culture's position grow more precise, our capacity to predict its trajectory degrades, because the machinery of cultural description and cultural action have merged. We close with ethical questions raised by AI-mediated cultural drift, recursive synthetic culture, and the limits of human cultural agency, offering guidelines for safe, equitable, and privacy-preserving research on AI and culture.

We investigate the use of formal methods to provide tight and sound generalization bounds for learning algorithms. By casting the traditional notion of algorithmic stability as a specification to be verified, we demonstrate that recent advances in reachability analysis can yield provable bounds on the generalization of a given model and algorithm on a sample dataset. As sample-specific algorithmic stability is insufficient to bound the usual distributional notion of generalization, we develop a novel concentration inequality to connect the sample-specific results of formal certification algorithms to the required distributional analysis for bounding the expected generalization gap. The resulting framework enables the analysis of prior generalization bounds to extend far beyond their original restrictive assumptions. Our approach computes sound bounds on the expected generalization gap in a constant number of algorithm runs without making any analytical assumptions on the algorithm; to achieve non-vacuous bounds we only require that the certified reachable parameter set is bounded --- a condition that we do not assume but formally verify. In practice, we demonstrate that our framework provides formal generalization guarantees that are orders of magnitude tighter than alternative sound computational approaches at scales ranging from toy datasets to fine-tuning classification heads on top of modern large language models. While we implement certification-enhanced versions of several well-known stability results, future extensions of our approach will enable tighter bounds and enhanced practical adoption across the spectrum of modern generalization bounds.

Token-level differentially private (DP) prompt optimization methods such as DP-OPT can become unstable under tight privacy budgets: on GSM8K, DP-OPT obtains $49.5\pm28.5\%$ across 30 runs, and a logged search trajectory reveals prompt-template drift and noise-sensitive irreversible choices. We diagnose these as structural consequences of greedy token-by-token construction over privately aggregated counts. We then propose DP-ES (Differentially Private Evolution Strategies), a structurally cleaner alternative that maintains a population of full prompts, mutates them via LLM calls that never access the private dataset, and spends privacy only on sampled-Gaussian evaluation; deterministic or Gumbel-smoothed selection is post-processing. Under a conservative $(\varepsilon\leq1.0,δ=10^{-5})$ guarantee, DP-ES achieves 88.1% on GSM8K (+38.6 pp over DP-OPT, approximately 9 times lower standard deviation), 99.7% on MedQA, 73.5% on BANKING77, and 86.8% on Alpaca. It is also 2.5 times faster in wall-clock time and uses 3.3 times fewer logged private-data call groups than DP-OPT. Selection and population ablations, implementation-level noise checks, and a 200-profile exact-match memorization stress test complement the formal guarantee. Scope: Our experiments establish optimization robustness under DP noise, especially where prompt structure is critical; end-to-end validation on genuinely sensitive, non-saturated deployment data remains future work.

Instruction following is central to language-conditioned robot policies: language should determine what to do when the same scene permits multiple valid actions. Yet successful execution alone cannot establish whether a policy follows the instruction or infers the task from the scene. We study this ambiguity through scene-preserving instruction interventions, using valid target substitutions, arbitrary nouns, and unrelated sentences while holding the scene fixed. We evaluate vision-language-action (VLA) policies and world-action models (WAMs) in simulation and in real-world experiments. Our analysis addresses three questions: (a) Does task success imply instruction following? When instructions request a different visible object, all evaluated policies predominantly approach and pick up the incorrect original target associated with the scene. (b) Is this failure caused by language insensitivity? Instruction perturbations affect task performance. A layerwise action lens shows intermediate action predictions respond to these perturbations. Linear probes accurately recover instructed targets, indicating modified instructions are encoded despite rarely determining target selection. (c) Why does encoded language fail to control action? Attention analysis indicates weak instruction-token contributions to action generation. Target-token attention can remain focused on the original object, revealing a mismatch between target encoding and visual grounding. UMAP and shared non-negative matrix factorization show target information remains accessible within representations increasingly organized by scene identity. Our findings expose a grounding gap concealed by nominal success and provide a diagnostic framework. They further establish a concrete criterion for progress: policies should reliably follow valid changes in user intent, even when they conflict with scene-favored behavior.

Multi-instance partial-label learning (MIPL) addresses inexact supervision in both the instance and label spaces, which can be applied to video classification. However, bag-level labels do not explicitly supervise the correspondence between candidate classes and temporal evidence. We propose {\ours}, which couples label disambiguation with temporal evidence allocation through a joint class--time assignment. Occupancy-regularized spherical matching associates contextualized video features while learning nonuniform temporal mass and discouraging excessive concentration. During training, candidate-restricted inference recomputes the assignment within the candidate label set. A dual-marginal KL projection then constructs a structured teacher that incorporates momentum-refined class beliefs while preserving the proposal's temporal occupancy. A single plan-level KL objective aligns the full-space predictor with this teacher. Our analysis characterizes when candidate re-solving differs from masking and shows that, under the stated construction, the joint objective decomposes into class-marginal and class-conditional temporal supervision. We construct VCMIPL benchmarks from Breakfast, DoTA, and FineAction using model-generated candidate labels and evaluate the method across four feature representations. Extensive experimental results demonstrate that PIVOTMIPL outperforms existing MIPL algorithms in both effectiveness and efficiency.

Self driving vehicles have emerged as a reliable technology that has the capability to transform transportation and mobility. The development of self driving cars requires significant advances in a number of areas, including perception, localization, decision making, and control. This research paper is based on the project implementation of the combination of object detection using YOLO (You Only Look Once), depth sensing using MiDaS for the localization and perception of obstacles, perspective transform, and decision making for path planning in self driving cars. The contemporary state of the technology for object detection, depth sensing, localization, and path planning evaluates the performance of the combined system through simulations and experiments. The results show that the combination of YOLO and MiDaS provides a new robust system for object detection and depth sensing. This research paper contributes to the advancement of self driving car technology and provides new and innovative approaches to the perception and localization of obstacles in the environment. Keywords: YOLO, MiDaS, perception, localization, decision making

A multi-domain eXtended Physics-Informed Neural Network (XPINN) framework is developed for nonsmooth Delay Differential Equations (DDEs). This is the first implementation to demonstrate the efficacy of partitioning the temporal domain into subdomains of integer multiples of the characteristic time delay and progressively training the associated subnetworks while freezing previously learned parameters. The efficacy of the proposed framework is demonstrated using a machining dynamics model that incorporates both regenerative and nonsmooth frictional effects. Results demonstrate that the proposed multi-domain XPINN framework leads to better solution reconstruction in DDEs and improved parameter estimation compared to a generic PINN (SPINN) formulation. The proposed method works particularly well for extended temporal domains and non-constant history functions. The robustness of inverse XPINN (I-XPINN) is also assessed using reference data contaminated with Gaussian measurement noise. Results indicate that I-XPINN remains resilient to measurement noise and the physics-informed constraints guide the network toward accurately recovering the underlying dynamics. This demonstrates, for the first time, the potential of the proposed framework for reliable parameter identification in DDEs characterised by nonsmoothness and large time delays.

Spherical data arise in many scientific applications. Often spherical transformers disregard the geometry of the underlying spherical domain, causing distortions and coordinate singularities near the poles. We introduce SO(3)-RoPE, a relative positional embedding that incorporates spherical geometry into transformer attention through unitary SO(3) representations. Our formulation is SO(3)-equivariant and compatible with FlashAttention, retaining efficiency of vanilla transformers. On shallow water dynamics prediction over a rotating sphere, our SO3ViT outperforms an S2Transformer baseline with lower errors and reduced runtime.

Establishing ordinary differential equations (ODEs) describing population data is a fundamental part of mathematical modeling in pharmacology, crucial to developing digital twins. However, doing so from sparse, noisy data is a slow, expert-driven task. Existing automated methods either search a restricted model space or ignore population inter-individual variability. Here we introduce APOD (Agentic Population ODE Discovery), a language-model agent that iteratively reasons over biological knowledge and fit diagnostics in an open-ended search space to discover a population digital twin (PDT), i.e., a shared ODE system with between-subject variability. On synthetic pharmacokinetic and tumor-dynamics benchmarks, APOD recovered ground-truth structures in 94-100\% of runs, 12-fold faster in median than an established library-based search. On real cohorts it converged to valid structures, and proposed a PDT of radioligand-therapy-induced platelet dynamics that predicts thrombocytopenia from first-cycle data and simulates alternative dosing schedules that lower the predicted risk of toxicity.

Prior audio-visual self-supervised learning methods rely on mechanisms such as EMA target encoders, prediction heads, reconstruction decoders, and contrastive losses. We introduce LeAVJEPA, the first audio-visual encoder trained under LeJEPA's collapse-free objective. A single early-fusion Vision Transformer processes audio, video, and joint audio-video inputs. Modality dropout treats a missing modality as another view of the same event, making cross-modal alignment implicit in the objective. The model aligns global embeddings with modality-specific local embeddings, and SIGReg prevents representational collapse. A controlled ablation identifies modality dropout as the key mechanism for audio-visual alignment. Despite the architectural simplicity, LeAVJEPA reaches 36.0 mAP on AudioSet-20K and 91.3% accuracy on ESC-50 under frozen evaluation. After fine-tuning, it reaches 61.1% accuracy on VGGSound, and its embeddings support zero-shot audio-visual retrieval.

Pre-generation success probes estimate response correctness from a language model's hidden activations before decoding, enabling cost-aware routing. While prior work has demonstrated their utility primarily on English inputs, we study their reliability across languages along three dimensions: (1) whether they preserve the ranking of likely successes and failures (DISCRIMINATION); (2) whether they retain probabilities that match observed success frequencies (CALIBRATION); and (3) whether they produce scores comparable enough across candidate models for cost-aware multilingual routing (UTILITY). Using 3,000 MATH problems in 10 languages and 8 open-weight model configurations, we compare cross-lingual transfer from English-trained probes and equal-budget pooled multilingual probes. English-trained probes retain useful cross-lingual discrimination but become less well calibrated after transfer. Pooled multilingual supervision improves both properties and yields more reliable estimates of success. In routing experiments, the pooled router achieves a 0.7% higher test success rate while reducing modeled cost by 13.0% relative to always selecting the model with the highest average success. These results show that multilingual routing requires success estimates that remain well calibrated and comparable across languages and models.

We study the sampling allocation of LinUCB in the small-gap regime, where the reward gaps are of order at most $n^{-1/2}$ over the decision horizon $n$. This scaling captures the hard instances underlying worst-case regret lower bounds, for which LinUCB is known to be near optimal up to logarithmic factors in $n$. Using a mean-field perspective, we characterize this allocation through the empirical sampling distribution, a macroscopic object that averages the effect of adaptive decisions over the horizon, and identify its limit as $n\to\infty$. We establish that in this regime, the empirical sampling distribution induced by LinUCB converges to the set of D-optimal designs. This central result reveals that, in the small-gap regime, LinUCB not only achieves near optimal minimax regret but also allocates samples in a way that is asymptotically efficient for learning the reward parameter, thereby connecting regret-driven online learning with information-efficient experimental design. Building on the optimal design limit, we obtain two useful consequences. First, we refine the asymptotic regret analysis of LinUCB in the small-gap regime by characterizing its leading-order constant in the limit. Second, we show that, despite LinUCB's adaptive sampling strategy, the regularized least-squares estimator satisfies a central-limit-type theorem in the small-gap regime, thereby enabling valid statistical inference for the reward parameter.

Learned models for industrial control are usually judged by aggregate accuracy, but accuracy at the component level does not guarantee safety once it is embedded in the system it is meant to serve. We study this gap on a behavior-cloning task: imitating an expert Nonlinear Model Predictive Control (NMPC) policy for load-following of a Pressurized Water Reactor (PWR), an industrial system with tight safety constraints. We propose a structured architecture encoding variables from each timescale into separate latent spaces, reflecting the physical decomposition of the system, before training a controller to imitate the expert on the product latent space. On long-horizon rollouts, separated embeddings improve both accuracy and feasibility compared with a shared-embedding baseline. Sensitivity analysis further shows that our model yields interpretable representations aligned with the system's physics. However, standalone deployment still leaves several percent of trajectories infeasible regardless of the architecture. Using our method to warmstart the NMPC optimizer rather than acting standalone, we recover full feasibility and near-optimal cost while still cutting computation time by $\sim$15% relative to the expert controller, and even more for abrupt operating changes.

Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and AI agents are increasingly embedded in operational business processes. Yet Business Process Management (BPM) curricula and frameworks still largely treat artificial intelligence (AI) as an add-on technology, leaving graduates (as potential future process developers) unprepared to reason about AI as a first-class design element of end-to-end processes. This paper addresses that gap by proposing \emph{AI-decision checkpoints}: explicit moments in a process development trajectory where process developers identify AI-candidate sub-processes, assess expected effects on time, cost, quality, and flexibility, consider legal and organisational constraints, and document a reasoned decision to adopt, constrain, or reject specific AI components. The checkpoints are instantiated through a fictitious customer onboarding process as a \textit{BPM Teaching Case}, embedded in a lifecycle-driven framework spanning six modules that combine process modeling, simulation, workflow execution with AI agents, and process mining, with each module's output serving as the next module's input. A preliminary formative reflection draws on instructor observations, submitted artifacts, and discovered process maps from learning-management-system logs. These exploratory observations suggest that the approach supported clearer distinctions between task-level automation and process-level value.

LLM agents are starting to own the full customer experience. Soon, LLMs may be selling and buying on behalf of companies and customers respectively. Small models are more cost-efficient at scale, but can reinforcement learning train them into competent sellers? We train four Gemma 4 checkpoints (2.3B to 31B effective parameters) with GRPO on a programmatic utility reward for bilateral multi-issue bargaining, and evaluate every arm on the same 1,152 negotiations against two frontier buyers it never saw in training. With the same learning rate ($10^{-6}$) for every size, the gain of the RL model over its base rises from $+0.001$ at 2.3B to $+0.078$ at 31B. Each size was trained once and the two smallest checkpoints use a different architecture, so we fit no scaling law. Tripling the learning rate, with the same or fewer training steps, improves on the shared rate at every size by $+0.032$ (2.3B) to $+0.081$ (4.5B). In exploratory comparisons with two frontier models run as sellers, the 12B seller trained at the tripled rate scores above both, though its untrained base already scores as high as they do. The 4.5B seller at that rate shows no detectable difference from either and fits on one 48 GB GPU. A further 2.3B arm at ten times the shared rate raises pooled score, but its gain concentrates on the evaluation buyer that shares a model family with the training pool. These results suggest tuning the learning rate before concluding that a small model cannot learn to negotiate, and testing against buyers from more than one model family.

Learned feature representations in neural networks often contain structure beyond that directly used by the final task output. We study this structure using $\textit{passive probes}$ that apply fixed, untrained, property-independent projections to representations as they evolve during training. We motivate this approach through the task of prediction on $S^2$ where equivalent vector and Hermitian parameterizations reveal an additional loss-invariant trace coordinate. This motivates a general construction in which fixed random projections serve as observers of learned features. Because the observer is loss-invariant and independent of the property being studied, changes in accessibility reflect changes in the representation relative to the fixed observer rather than adaptation of the observer itself. We show that ensembles of passive probes can directly reflect task-relevant information such as target alignment. Under our constructions, the accessibility of eventual difficulty evolves differently across tasks. It increases during training in the regression tasks of surface-normal estimation and image inpainting but remains near its initial level in image classification. Comparisons with learned linear probes further show that recoverability and passive accessibility can evolve differently during training. Together, these results show how passive probes can separately characterize changes in representation geometry and the accessibility of eventual task difficulty.

Autonomous coding agents now resolve a substantial share of real-world GitHub issues. However, passing functional tests differs fundamentally from producing a high-quality contribution acceptable for merging. Mature open-source projects publish repository-specific contribution policies, spanning style, git, testing workflows, to ensure code quality and long-term maintainability. Because existing benchmarks evaluate patches solely on unit tests, agent compliance with repository governance remains unknown. In this paper, we introduce SWE-CC, a benchmark evaluating code and process compliance in autonomous software engineering. We develop a semi-automated pipeline that converts developer documentation across 12 open-source repositories into 823 machine-checkable atomic policies. SWE-CC introduces two features: 1) lightweight, deterministic checker functions that represent each policy, 2) a comprehensive auditing mechanism that inspects both agent runtime behaviors and final deliverables. We evaluate the compliance of agent workflows in 500 end-to-end software contribution tasks extended from SWE-bench Verified. Our evaluation of four LLMs under two agent scaffolds shows that modern agents suffer from coding compliance issues: although agents produce functionally correct patches, they still violate 43.1 percent of applicable project policies, with nearly half of all violations occurring during intermediate execution steps. These results show that functional correctness does not guarantee real-world readiness, highlighting that future software engineering agents must reliably conform to repository governance to enable safe and trustworthy deployment.

Multi-agent systems built on large language models (LLMs) restate observations as a matter of course: relays forward them, shared boards repeat them and discussion rounds echo them. An aggregator that pools such messages should count sources, not statements. We convert a reported probability into units of independent readings, which assigns every restatement a copy weight, 0 for an aggregator that counts sources and 1 for one that counts every statement, and yields the implied decision under any cost structure. Three testbeds hold the evidence fixed and vary how it is restated: message logs with an exact Bayesian oracle, web documents with appended copies, and logs written by LLM agent teams under four communication protocols. Across four models from three providers, a forwarded copy counts for 0.06 to 0.42 of a new reading, mostly because some replies count every statement. On 5% to 40% of logs that state one reading three times, the reported belief implies an early commitment that the oracle never makes. The models that count copies least and most on controlled logs do so on web copies and agent-written logs as well. A one-paragraph declaration of what a copy contributes brings the copy weight on controlled logs to 0.08 or less. A rule that has agents refer to readings instead of restating them cuts belief-implied early commitment from 11.2% to 1.1% and preserves genuine corroboration.

An agent whose tool keeps returning nothing useful should stop relying on it. In a retrieval environment with controlled source failures, we separate how agents judge results from what they do. We compare stopping at the same step after longer and shorter runs of results the agent judged useless; this contrast is zero for clock- or deadline-driven stopping. Where we record their judgments, the seven agents we test call a failing source's results useless 97-100% of the time, yet most of them rarely stop on that judgment. Prompt cues change when they stop but not what they stop on. Permission to answer from memory and a reasoning mode can bring early stops regardless of evidence, a stated budget moves the 7-8B models' stops to the deadline, and a stopping rule or call cost in the prompt is followed at most partly. Stopping follows the evidence only when the harness enforces an integration step that makes the agent answer after five consecutive results it judged useless. This step raises failing-source success for every model, keeps the stopping point fixed when the budget doubles, and needs no extra judgment call when the agent states its judgments. A pre-registered replication on 300 fresh questions confirms the dissociation and the rule's effect.

Large language models (LLMs) can reason over scientific literature to devise design strategies, yet fail to reliably implement them for biological sequences. While protein generative models learn sequence patterns, they lack the capacity to incorporate literature evidence for multi-step reflective reasoning, forming an evidence-to-execution gap between scientific reasoning and sequence manipulation. We present EASER (Evidence-Aware Sequence Engineering with Reflection), a reflective agent bridging reasoning and sequence generation via a learned property interface of offline-trained, fixed low-rank matrices. The agent steers a diffusion generator by combining these matrices, proposing intervention hypotheses (anchors, editable positions, control coefficients) grounded in retrieved evidence, sequence context and past results. A Probe-and-Steer mechanism validates interventions and allocates samples according to predicted property responses, with outcome reflection informing subsequent decisions. Evaluated on multi-objective antimicrobial peptide design (optimizing activity, non-hemolysis and non-toxicity), explicit hypothesis formulation delivers better multi-objective performance than direct action generation under identical decision conditions. Ablation studies verify the importance of evidence retrieval, episodic history, reflection and Probe-and-Steer. Over six repeated trials, EASER obtains the highest mean hypervolume and lowest mean IGD+ on screened candidates compared with competing baselines. Our work demonstrates how an executable property interface and iterative feedback link scientific reasoning to targeted peptide sequence generation.

Lloyd's algorithm and the discrete local (D-local) optimization method (Li et al., 2025) for $k$-means provide only weak local-optimality guarantees, and their solution quality remains sensitive to initialization. In this paper, we introduce $r$-point local optimality, under which no reassignment of at most $r$ samples decreases the objective function, and focus on $r=2$. The main computational obstacle is the $\mathcal{O}(n^2(k^2+d))$ cost of exhaustive two-point certification for $n$ samples in $d$ dimensions and $k$ clusters. To address this challenge, we prove that (i) every improving two-point move of a D-local optimum must involve a cluster shared by both reassignments, and (ii) only certificate-defined boundary points can participate in an improving pair. Exploiting this structure, we propose Boundary-Point-Screened Two-Point Local Search (BPS-2PLS), which terminates at a two-point local optimum. For fixed $k,d$ and nonvanishing cluster occupancy, the number $m$ of retained candidates satisfies $m=\mathcal{O}_{\mathbb{P}}(\log n)$ under i.i.d. sampling from a bounded-support distribution with bounded density or from a Gaussian mixture. Across twelve benchmarks, BPS-2PLS attains the lowest available mean WCSS on ten. In a subsampling study, screening retains 0.10% to 2.81% of samples on average at the largest tested sizes. The code is available at https://github.com/lwl-learning/BPS-2PLS.

Whether a time-series pattern is anomalous often depends on the operating regime of the monitored process. A missing event can signal a fault in one regime and be routine in another, and the query alone may not reveal which regime applies. We study in-context learning (ICL) for time series anomaly detection (TSAD) through reference-conditioned detection, where a reference record provides evidence about expected behavior and model parameters remain fixed at inference. Supplying the reference is not enough: when training anomalies are recognizable from the query alone, the detector can fit its targets while ignoring the reference. We therefore introduce counterfactual supervision, which pairs one query with two references that support different normal rules and labels the query under each. At positions where the two labels disagree, no detector that ignores the reference can fit both targets. Anlu learns from this supervision by adding a reference memory and zero-initialized gated adapters to a frozen time-series foundation model (TSFM) pretrained for anomaly detection. On the 350 TSB-AD-U evaluation sequences, Anlu raises the mean VUS-PR of the frozen TSFM from 0.542 to 0.607. Replacing the reference with zeros lowers Anlu's score to 0.499.

A patient-timeline reconstruction system is auditable only if it keeps the mentions behind each answer, records how facts were revised, and declines to answer when the evidence is not in the text. This study tests these three properties on a fully synthetic corpus (1,000 patients, 3,353 notes, 220 revision edges). Two provenance-aware Evidence Graph operators reduced the node-plus-edge count to 67% and 63% (77-78% of serialized size) while preserving every answer and mention link across 6,813 query points answerable by recency; a fixed-window baseline returned no value for 53.4% of points, unflagged. On evidence-unavailable controls that announce the omission, a BioClinicalBERT gate and a zero-shot LLM gate responded mainly to the announcement. On marker-free controls, BERT abstained on 0 of 81 notes while its accuracy fell from 93.8% to 59.3% across all three relation classes; the LLM's coverage fell from 75.6% to 27.7% on notes its own model family judged undeterminable. Against 482 regenerated gold spans, the LLM's cited evidence reached recall 0.850 and precision 0.864; BERT's span head, trained without span labels, did not localize evidence. A temporally versioned provenance graph stored abstentions as typed, queryable edges. The clean task admits a 0.651-accuracy shortcut, and results describe implementation behaviour on synthetic data, not clinical performance.

Near-term quantum natural language processing (QNLP) experiments often run on classical simulators, so simulator cost is part of the field's practical compute burden, yet accuracy tables do not show it. We measure that cost for a variational quantum classifier (VQC) on binary SST-2 sentiment classification, using PennyLane's state-vector simulator over a controlled grid of 27 configurations: three balanced training-set sizes (N = 200, 500, 1000), three qubit counts (4, 6, 8), and three circuit depths. Each VQC is compared with logistic regression on the same PCA-reduced input; full TF-IDF logistic regression gives an uncompressed reference. The VQC beats its matched baseline in 6 of 27 single-seed comparisons. After reruns at two further seeds, only 1 of these 6 keeps a positive mean advantage larger than its paired seed-to-seed variability, and paired tests on the fixed validation set do not establish it. VQC training is 886-21,127 times slower in measured wall-clock time than the matched classical fit (median 3,158 times); going from 4 to 8 qubits roughly doubles simulator time, and within the tested range per-step cost is well approximated by a linear function of the parameter count. CodeCarbon energy and CO2 estimates are secondary: they imply an almost constant power of about 41 W, so they add little beyond runtime, and we do not build an energy ratio from them. The study is narrow (one dataset, representation, ansatz, simulator, and CPU environment) and is a reproducible feasibility measurement, not a general verdict on QNLP.

Can LLMs recognize degradation in their own computational substrate? Inspired by anosognosia, a neurological condition in which patients fail to recognize impairments in their own abilities, we investigate whether LLMs can recognize degradation in their computational substrate induced by quantization. We first show that existing models fail to self-report their quantization state, even when provided with their own generated text as an external cue. Linear probing reveals that, while generated text carries almost no trace of quantization, internal representations contain clear, method-specific fingerprints. Through training, models learn to identify severely degraded outputs such as those of 4-bit models by comparison, yet still fail to do so from a single output. A shared LoRA trained jointly across quantization levels succeeded in reading out internal fingerprints, but fails on unseen quantization methods, merely mapping method-specific fingerprints to labels. Whereas external self-observation can restore awareness in some cases of human anosognosia, our results suggest that the more promising route to enabling such awareness in LLMs may lie in their internal representations. Our results highlight fundamental limits of generalizability to LLM self-monitoring.

Biomedical large language model (LLM) evaluation requires auditable assessment of narrow, evolving, source-grounded subspecialty knowledge. Multiple sclerosis MRI (MS-MRI) provides a high-stakes textual-knowledge test case because correct reasoning requires current diagnostic criteria, standardized acquisition and reporting knowledge, longitudinal monitoring concepts, lesion morphology, and recognition of difficult mimics. We present MS-Exam-Gen, a reproducible framework for constructing and auditing a text-based multiple-choice question (MCQ) benchmark for MS-MRI knowledge; it does not evaluate direct MRI image interpretation. MS-Exam-Gen targets source-grounded criteria, protocols, reporting, and differential diagnosis. The framework combines expert-source indexing, exam-oriented topic induction, evidence-grounded MCQ generation, automated quality audits, a same-family consistency screen, and empirical calibration. From a 66-source corpus indexed into 4,289 retrieval chunks, the pipeline produced a locked 3,058-item candidate benchmark spanning 16 topics and 53 subtopics. Evaluation across 12 primary LLM endpoints yielded 36,696 item-level predictions and separated performance over a 42.8-percentage-point accuracy range (89.7% to 46.9%). Across these endpoints, 25.5% of items were missed by at least four. Post-generation audits showed that refreshed construction reduced measurable answer cues, while option-order testing showed that absolute MCQ scores remain position-sensitive. Generated construction labels remain metadata rather than validated psychometric categories. Because expert adjudication and full option-order counterbalancing remain future work, MS-Exam-Gen is not a clinically certified examination. It should be interpreted as an automatically filtered, source-grounded candidate benchmark and reproducible audit workflow for item-level and topic-specific LLM evaluation.

Localizing where an LLM-based agent first fails to uphold security during a repair can show which stage of its workflow needs an additional safeguard. This is difficult for silent failures, which are patches that pass syntactic and functional checks but still contain a security vulnerability. Because such patches give no observable failure signal, existing failure attribution methods, which rely on observed task failures and labelled failure steps, are less suited to them. We propose Security Awareness Gap Evaluation (SAGE), a trace-based method that combines an assessment of the security reasoning recorded at each turn with the reconstructed code history to identify the earliest turn at which a repair diverges from the task's security intent. We evaluate SAGE on 95 confirmed silent failures drawn from 3,684 repair traces produced by six agent frameworks and six base models on SecurityEval and CVEfixes. SAGE assigned an origin in 93 cases. Most origins were an unaddressed security requirement or an inadequate defence choice, and only five coincided with the code change itself. When the agent introduced the vulnerable code, the origin preceded the write in 14 of 19 cases. Repeated scoring and a second judge reproduced the origin type more consistently than the exact turn, and agreement was lowest for traces that kept only the final file.

During language acquisition, bilingual children are regularly exposed to code-switched input and use it as a cognitive scaffold to accelerate vocabulary growth and cross-linguistic syntactic mapping. In contrast, computational bilingual models are conventionally pretrained on interleaved monolingual corpora. While introducing synthetic code-switching during pretraining has become a promising strategy to enhance cross-lingual alignment and downstream performance, the structural and developmental parameters governing the success remain poorly understood. In this work, we investigate the efficiency of training with synthetic code-switched data across two typologically distinct language pairs by controlling two key variables: the structural location of code-switches and the dynamic switching rate across training stages. Our results show that training with code-switched data improves cross-lingual alignment for typologically close languages.

A representation family is a distinct way of extracting features from time series. Ensemble algorithms that combine several representation families remain the most accurate approach to time series classification. Current state-of-the-art ensembles, most notably HIVE-COTE~2.0, pair a bespoke classification algorithm with each representation family and combine their predictions using a fixed, non-adaptive rule. We present TIGER (Time-series classification with In-context-learning Gated Ensemble of Representations), which instead applies the same small portfolio of three general-purpose classifiers (Ridge, Extra Trees, and Naive Bayes) to four representations from four distinct families, stacking the resulting twelve base learners' predictions into a meta-feature matrix. The final prediction is produced by an adaptive meta-classification rule that chooses, independently for each data set, between a weighted hard majority vote and TabICLv2, a pretrained tabular foundation model used in-context as a meta-classifier, based on the mean number of training samples available per class. On a 142-data-set benchmark drawn from the UCR time series classification archive, TIGER obtains the best mean accuracy, balanced accuracy, and F1-score among six compared algorithms, including HIVE-COTE~2.0, and significantly outperforms each of the other five individually. TIGER's adaptive rule also meaningfully outperforms either of its two constituent meta-classification methods used alone, and its single hyperparameter, tuned using only a twenty-data-set development subset, is shown to generalize to the full evaluation benchmark. We further characterize TIGER's design through an extensive set of ablation experiments and report the design alternatives that we investigated and ultimately discarded.

Power Delivery Networks (PDNs) are critical components of modern VLSI chips, providing stable voltage levels while satisfying electromigration (EM) and IR-drop constraints. Conventional PDN design methodologies typically rely on worst-case assumptions, often resulting in over-provisioned networks and inefficient use of resources. This paper presents a reinforcement learning-based framework for the optimization of workload-aware PDNs. The proposed methodology first generates workload-aware PDNs using architectural power traces obtained from system-level simulations. These power traces are mapped to spatial power density distributions, enabling adaptive allocation of PDN resources according to local current demand. A reinforcement learning agent then performs wire-width optimization to minimize PDN area while maintaining EM and voltage integrity constraints. Electrical and reliability metrics are obtained using SPICE-based circuit analysis and EM lifetime estimation. Experimental evaluation is performed on a dataset of workload-aware PDNs generated from 4-, 8-, and 16-core multiprocessor floorplans using PARSEC and SPLASH-2 benchmark workloads. Furthermore, the proposed Deep Q-Network (DQN)-based optimizer reduces the average normalized PDN area by 47\% while satisfying all EM and IR-drop constraints. Compared to simulated annealing, the proposed approach achieves comparable optimization quality while providing approximately 26$\times$ faster optimization.

Vision-language models (VLMs) achieve strong zero-shot transferability but remain vulnerable to target-domain shifts at inference time. Test-time adaptation (TTA) offers a practical remedy, yet most existing VLM-TTA methods follow a prediction-side adaptation paradigm. They use test samples to adjust logits, prototypes, caches, priors, or feature statistics, often incurring additional computational overhead. In this paper, we take a different perspective and reframe VLM-TTA as candidate verification rather than prediction adjustment. We propose Test-Time Correction (TTC), a hypothesis-based correction framework guided by a simple principle: hypothesize, reconstruct, correct. Given a test feature and its top-k candidate labels, TTC treats each candidate label as a hypothesis, reconstructs the feature within the corresponding latent subspace stored in a memory bank, and measures the resulting divergence shift. This shift quantifies how much the candidate subspace and its relations to other candidates change after the hypothetical insertion of the test feature. A correct candidate hypothesis induces only a small shift, whereas an incorrect one perturbs the subspace more strongly. TTC therefore corrects the prediction by selecting the candidate with the minimum aggregated divergence shift. This training-free candidate-verification mechanism avoids iterative optimization and provides a favorable accuracy-efficiency trade-off. Across five TTA settings and 15 benchmark datasets, including zero-shot classification, domain generalization, few-shot classification, base-to-novel generalization, and cross-dataset evaluation, TTC consistently improves accuracy over state-of-the-art VLM-TTA methods while achieving up to 2x speedup, over 3x lower CPU memory usage, and up to 1.4x lower GPU memory usage than the lowest-memory training-free baseline.

Accurately quantifying the predictive uncertainty or improving model calibration plays an important role in medical image classification, in particular in melanoma diagnosis, where accurate uncertainty quantification can have significant implications for patient care. One of the methods for calibration improvement is data augmentation. In addition, data augmentation as a method for synthetically increasing the size of the dataset has been proven to improve the performance of models trained on imbalanced datasets. However, the impact of data augmentation, as a transformation of a part of the original data, on calibration of models trained on imbalanced datasets, in particular in melanoma classification is under-explored. We train neural networks on SIIM-ISIC 2020 melanoma classification dataset under two conditions: with and without data augmentation, and compare the differences in AUC and expected calibration error (ECE) in both scenarios. Our results shows improvements in uncertainty calibration using different augmentation methods.

Quantum compilers synthesize high-level operations, such as the quantum Fourier transform and multi-controlled X gates, into gate-level circuits. Across compilation stages, a circuit may be serialized, exchanged as OpenQASM, converted between compilers, or lowered to gates. These representation changes can preserve computation while removing the high-level operation itself, leaving the receiving compiler unable to apply a requested synthesis method. We call this loss synthesis availability. We study synthesis availability in Qiskit, TKET, and Cirq through controlled experiments, repository analysis, circuits captured from project tests, and a Munich Quantum Toolkit pipeline. Synthesis availability survives serialization when deserialization restores the high-level operation and survives compiler conversion only when the receiving compiler supports that operation. It does not survive the evaluated OpenQASM routes. After a direct OpenQASM 3 round trip, requested synthesis has no effect in all 360 delivered conditions even though compilation succeeds. In the MQT pipeline, OpenQASM 2 interchange increases the two-qubit gate count by 37.2 percent for an eight-qubit Grover circuit. We also introduce Verified, which records high-level operations before a representation change and reconstructs them only when a check confirms that the record still matches the delivered circuit. Verified restores requested synthesis in all 830 recovery runs, with median verification times from 2.0 to 638 ms per operation family. The check rejects every stale record but also some valid records. These results show that preserving functional equivalence alone is insufficient when later synthesis choices depend on retained high-level structure.

Melanoma is the deadliest type of skin cancer, whose early diagnosis is crucial for patients' survival. Image classification using deep learning models has shown promising results for melanoma diagnosis. However, the performance of these models on the melanoma datasets such as SIIM-ISIC melanoma classification dataset is a challenge due to the class imbalance. One of the methods to deal with this challenge is using loss function modifications. In this work, we have investigated the effect of different loss functions on the performance of deep neural networks. We trained these networks using focal loss, logit-adjusted softmax cross-entropy (CE) loss, and weighted softmax CE loss, and we report different metrics for evaluating performance and uncertainty calibration. Our results suggest that focal loss delivers a good combination of performance in terms of AUC and uncertainty calibration in terms of expected calibration error (ECE) simultaneously.

Graph Convolutional Networks (GCNs) have emerged as a powerful framework for learning from graph-structured data, yet their deployment on resource-constrained edge platforms remains challenging due to the computational and memory demands of sparse graph aggregation. This work presents an FPGA-based GCN accelerator that combines DSpar graph sparsification, 8-bit quantization, and approximate multipliers on the AMD Kria KV260. Evaluated on Cora, LastFM Asia, and Amazon Photo, the design explores the interaction between sparsification and approximation across graphs with widely varying densities. Results show that the effectiveness of approximate arithmetic is governed by accumulation depth within GCN computations. Approximate multipliers are most effective when applied to sparse aggregation operations, while graph sparsification further improves their viability by reducing aggregation depth. The combined approach achieves up to 9.88$\times$ speedup while maintaining 86.6\% classification accuracy on Amazon Photo, and 1.52$\times$ speedup with 77.0\% accuracy on Cora, with total power consumption below 1 W. These results demonstrate that graph sparsification and approximate computing are complementary techniques whose co-optimization enables efficient low-power GCN inference on edge FPGA platforms.

Journal entry anomaly detectors are commonly evaluated on the full population with ROC-AUC, precision and recall, ignoring the review budget and which anomaly types are found. We propose a type-aware evaluation combining per-type recall, fair-share type recall (FSR), which caps each type's credit at its budget share, type coverage and first-hit rank. We evaluate nine unsupervised detectors, a supervised reference and feedback-driven Deep Semi-Supervised Anomaly Detection (DeepSAD) on four real client ledgers with injected typed anomalies and a public synthetic ledger. On the largest client ledger, principal component analysis (PCA), an autoencoder (AE) and a variational autoencoder (VAE) each place on average 98 anomalies among the first 100 postings, but at least 95.8 belong to one type. FSR instead favours a nearest-neighbour (kNN) detector and changes the top-ranked detector on three of four client ledgers. Representation also matters: one-hot encoding exposes unseen accounts, whereas frequency encoding leaves unseen contra accounts largely undetected. On the public ledger, the Histogram-Based Outlier Score (HBOS) and Empirical Cumulative Distribution-Based Outlier Detection (ECOD) reach all eight markings within 1,386 entries, whereas kNN, the hit leader at 1,000 entries, first reaches cross-linked clearing at rank 4,641, and the supervised row-level reference misses this marking within 1,000 entries. There, the adaptive DeepSAD review protocol raises mean hits per 100 reviews from 40.0 to 68.3 but type coverage only from 2.7 to 3.0. These findings show that high hit rates can conceal systematic blind spots and suggest that feedback can reinforce existing detection patterns without broadening anomaly coverage.

Implementing security features---functionalities that protect sensitive data or prevent malicious actions by attackers---is important for ensuring the security and integrity of software systems. Correctly implementing access control, cryptography, or other security features is challenging as they require substantial domain expertise, careful design, and custom implementation to integrate them within the software system. Previous work has thoroughly studied security awareness, perception, practices, and expertise via surveys, interviews, and experiments. While they have shown that the implementation of security features are often supported by security libraries and frameworks, they have also shown that developers rarely are security experts, and make mistakes that introduce vulnerabilities into software when using them. Despite this extensive knowledge of security practices and developer behavior, we still lack a comprehensive understanding of how security features actually manifest at the code level across full software systems. We close this gap by conducting a mining study of security features in 9 popular and large open-source Java systems. We manually inspected 2,127,761 lines of code across 19,121 files, identifying and annotating 183,395 lines implementing 561 security features corresponding to 54 security features in our taxonomy. We analyzed the characteristics of the identified security features, such as their size, scattering, tangling, common implementation patterns, and the use of internal and external functionalities. Our findings show that security features are far more than mere calls to external libraries. Security features, such as access control, can grow large in size, scatter across the whole codebase, and frequently tangle with each other. External security libraries are commonly used, but they require substantial amounts of code for integration.

Predictive maintenance requires reliable remaining useful life (RUL) estimation. Existing methods mainly follow two paradigms: wear-based aging models that capture cumulative degradation and sensor-driven data models that reflect instantaneous health conditions, each providing only partial information. In this work, we propose a probabilistic fusion framework that integrates wear-based and sensor-based prognostic components through failure probability distributions. Based on explicit structural assumptions linking wear, latent health, sensor observations, and failure, we derive a principled combination rule that enables uncertainty-aware integration with adaptive weighting of the components. Experimentally, we assess this combination rule by learning the wear-based component using a parametric survival model and the sensor-based component using a 1D convolutional neural network (1D-CNN) with a post-hoc uncertainty model. Evaluation on multiple N-CMAPSS datasets demonstrates that the fused model improves point accuracy, preserves the C-index, and produces narrower yet well-calibrated prediction intervals compared to either component alone. The results highlight the complementary roles of wear-based survival model and sensor-based deep learning model, and show that their probabilistic integration provides a structured pathway toward more robust and consistent prognostics over the life-time.

Embodied agents powered by large language models and vision-language models are increasingly deployed in physical environments, but jailbreak attacks can induce these agents to perform physically harmful actions. A growing number of guardrail methods have been proposed to intercept dangerous behavior before it is executed, yet existing safety benchmarks evaluate the embodied models themselves, leaving it unclear how well these guardrails actually defend an embodied agent in practice. We present the first systematic evaluation of jailbreak guardrails for embodied agents. To compare guardrails under identical conditions, we build a pluggable evaluation framework that treats the embodied agent as a fixed backend and each guardrail as a module that can intervene at the perception, planning, or control stage. We subject six representative guardrails to template-based and automated jailbreak attacks as well as safe instructions, and assess them at the system level along three dimensions: defense effectiveness, measured by the bypass rate and the hazard success rate in the simulator; usability, measured by the false-positive rate and the task completion rate on safe instructions; and efficiency, measured by the latency overhead added at runtime. Experiments on guardrails that span different intervention stages, decision mechanisms, and input modalities reveal a clear trade-off among the three dimensions, and show that no single guardrail dominates in all settings. We further analyze how intervention stage, decision mechanism, and input modality shape safety outcomes, and we offer practical guidance for selecting and designing guardrails for embodied agents.

Large learning rates can qualitatively change the trajectory of neural network training, often pushing optimization into regimes far from classical gradient-flow behavior. The Edge of Stability (EoS) offers a valuable lens on the dynamics such learning rates induce. We study corresponding dynamics in diagonal linear networks, where we uncover a competition between two distinct implicit biases that jointly determine the sparsity of the recovered solution in regression settings. Complementary to the Gain, which captures the average discretization error accumulated by Gradient Descent relative to Gradient Flow, we derive a closely associated but overlooked quantity: the Drift. Under large learning rates, it describes an imbalance between different discretization errors and represents a systematic shift in the optimization trajectory. While the Gain grows monotonically in certain regimes, and can bias towards denser, flatter interpolators, the impact of the Drift depends on its alignment with potential solutions, which can either counteract or reinforce the effect of the Gain. Consequently, its behavior drives model selection, particularly during early training epochs. To validate our theoretical insights, we introduce an intervention that actively steers the Gain to recover sharper, sparser solutions. Thus, our analysis reveals that large learning rates do not universally hinder the recovery of sparse solutions. On the contrary, they can be harnessed to control the implicit bias of training.

Modern LLM optimizers such as Muon often produce weight matrices with higher effective rank than Adam, yet further spectral control has delivered only modest gains. We identify a tension behind this result: concentrated spectra can suppress gradient directions in coupled weight matrices and slow optimization, while constraints maintained throughout training can limit task-specific adaptation and raise the attainable loss floor. We introduce ORCA (Orthogonal Regularization, Cooled After), a minimal optimizer intervention that applies strong but temporary soft orthogonality regularization early in training, then removes it. This allows the weights to benefit from a broader spectrum early on and adapt freely afterward. Across LLaMA, Qwen3, and fine-grained mixture-of-experts models ranging from 130M to 8B parameters, ORCA achieves lower final validation loss than Muon. Its loss reduction relative to Muon matches or exceeds Muon's reduction relative to Adam. Ablations support the early-shaping, later-release design. Further, ORCA requires no architectural changes and adds minimal overhead.

On-policy distillation (OPD) provides dense teacher supervision on student-generated trajectories, but generating and evaluating long rollouts incurs substantial training cost. Existing acceleration methods reduce this cost through open-loop rollout schedules or closed-loop horizon adaptation. However, supervision compatibility can vary substantially across trajectories, making a single rollout horizon difficult to match their heterogeneous reliable lengths: an overly short horizon may truncate useful supervision, while an overly long one wastes computation beyond reliable regions. Our key insight is that the trajectory-specific reliability boundary need not be predicted before generation. By viewing reliability as the first-passage of accumulated low teacher--student compatibility events, the boundary is inherently unknown before sampling, yet whether it has been reached can be determined exactly from the observed prefix. Building on this insight, we propose *Flash-OPD*, which shifts from rollout-horizon control to adaptive trajectory-level boundary verification. *Flash-OPD* interleaves cached generation with teacher verification and independently stops each trajectory according to its observed compatibility events. To reduce verification overhead, the recent event rate is used only to schedule the next verification point, while the actual stopping decision always relies on the exact cumulative count. This separation prevents estimation errors from causing premature termination while enabling efficient verification during generation. Extensive experiments across diverse datasets and teacher--student settings show that *Flash-OPD* achieves $2.2\times$--$7.5\times$ speedups over standard OPD while maintaining or improving accuracy.

Realistic environment replicas are increasingly valuable for training and evaluating LLM agents, yet the original systems may be inaccessible or impractical to reproduce. We explore agentic language world modeling: rather than rebuilding an executable environment, a world model agent serves as the environment for a task agent and supports faithful and stateful simulation. We instantiate this paradigm with Trace2Env, a learning-free framework for settings where the original system is unavailable but historical interaction traces remain accessible. Trace2Env reconstructs these traces into a reusable environment worldbook containing environment schemas, grounded evidence, and induced behavioral knowledge. At runtime, the world model agent actively consults the worldbook together with persistent episodic state to infer each action's observation and lasting state effects. Across nine environments, Trace2Env improves both next-observation fidelity and long-horizon interaction consistency over conventional prompt-based LWMs. In multi-turn interaction, task agent actions generated against Trace2Env remain valid more often when replayed in the real environment, indicating that its simulated dynamics better preserve the consequences of earlier actions across successive turns. These results establish agentic language world modeling as an alternative direction for building realistic environment replicas without reconstructing the original executable system.

In self-supervised learning, a third modality should improve, or at least preserve, performance. Across nine image-text-tabular datasets, we show that it instead harms performance: the trimodal model underperforms its own best bimodal subset in 55.6% of paired runs under VICReg. The same failure occurs in 51.1% of paired runs under SimSiam. We call this failure multimodal saturation. We propose that the failure lies in the alignment geometry. Riemannian VICReg (R-VICReg) generalizes classical VICReg: it aligns views by squared geodesic distance on learnable negative-curvature product factors and recovers VICReg exactly as curvature vanishes. Over the same 45 paired runs, R-VICReg raises the probability that the third modality helps from 44.4% to 64.4%, with gains concentrated where VICReg saturates.

Operator-learning benchmarks are stored at full precision and have grown to terabyte scale. Rate-distortion theory says how many bits the stored field needs, while a practitioner needs to know how accurate an operator trained on the compressed data will be. We show that the first does not determine the second, and measure why, compressing the input fields while targets and test inputs stay at full precision. A solution operator attenuates a perturbation of its input. Pushing a compressed field through a surrogate already trained at full precision measures how much of the perturbation that surrogate transmits. The fraction is consistent with the smoothing behaviour of the underlying equation, and it spans more than two orders of magnitude across PDE families. Field reconstruction error is computed before the attenuation and cannot see it. For operators trained with mean squared error it inverts 36 of 104 cost comparisons across datasets, where a probe built from the same forward passes inverts 12. Two families that PDEBench stores with identical initial conditions differ threefold downstream at identical field error. Under the relative-L2 objective of the reference recipe the separation narrows, while the ordering of the family-level median transmission factors is unchanged. After one full-precision training run, the probe evaluates an entire rate curve by forward passes alone. It ranks datasets and rates consistently across the codecs and architectures we test, while its magnitude does not transfer between them.

The robustness of Personally Identifiable Information (PII) protection in Large Language Models (LLMs) is a critical concern, yet the risks associated with cross-lingual data extraction remain under-explored. This study evaluates the vulnerability of English-centric and multilingual models to Training Data Extraction (TDE) attacks when prompted in non-English languages. We construct a multi-domain PII dataset comprising social media handles, email addresses, and phone numbers and translate the attack contexts into Italian, Spanish, French, and German. Our results show that TDE attacks against both English-centric and multilingual models transfer to different languages: the attacks are successful on translated prompts, even though only the original English prompt might have been included in the pre-training data. A web-presence check on a sample of the translations confirms that they are not available online. The share of English leaks recovered in other languages grows with the multilingual capability of the model, and it drops sharply when the original wording is lost, even without a change of language. This suggests that native multilingual pre-training facilitates the emergence of latent cross-linguistic bridges that simplify the retrieval of personally identifiable information (PII). We analyze the activations of multilingual large language models (LLMs) and find that different translations of the same prompt are bridged in similar representations, with the strongest alignment in the middle layers. Our results highlight a fundamental security gap in modern LLMs, necessitating more robust, language-agnostic sanitization strategies for future model alignment.

Diffusion- and flow-based robot policies have recently become widespread in robotic Imitation Learning (IL) due to their high performance and ability to model continuous and multimodal distributions. However, the iterative denoising procedure used by these models introduces significant prediction latency, hindering high-frequency closed-loop robot control and leading to jittery, unstable motion when frequent updates to the robot's action predictions are used. Therefore, it is common practice to train models to predict chunks of actions that can be executed sequentially without feedback, even when this reduces responsiveness and may mean the most recent state information is not used. In this article, we present Adaptive Mean Flow (AMF), a flow-based IL method that enables smooth and responsive, fully closed-loop robot control. AMF uses Mean Flow, which is an accelerated form of Flow Matching (FM), to minimize prediction latency. To ensure smoothness and consistency across predictions, AMF uses a corrupted version of the trajectory from the previous step when predicting new robot actions, with the signal-to-noise ratio increasing over the time parameter of the trajectory. This discourages large changes in the prediction from one step to the next, while allowing freedom to adapt the predictions for future steps. We evaluate AMF across a wide range of simulated and real robot tasks and demonstrate significantly improved performance compared with baselines. Code: https://github.com/akselva/Adaptive-mean-flow-RoboticIL.

Game-theoretic attribution explains a model by assigning credit to its features or training examples. The least core has attracted interest as an alternative to Shapley-style averaging because it can expose players that cause substantial harm in rare, high-value contexts. However, least-core allocations are generally nonunique, and the choice of allocation can affect the resulting explanation. In this study, we investigate how payoff selection and coalition sampling affect least-core attribution. Our experiments show that selector choice matters for distinguishing useful and harmful contributions, and that sampling can degrade harmful-player identification across the tested selectors even when useful players remain well identified. These observations motivate efficient computation with all coalition constraints and a well-defined selector. We introduce entropic least core (ELC), a smooth approximation whose unique minimizer follows a continuous path along the temperature to the nucleolus, a classical refinement of the least core. Our experiments show that ELC approximates the nucleolus faster than an LP-based nucleolus solver while retaining small payoff errors, with further GPU acceleration at larger problem sizes. In the tested full-coalition contextual-distractor games, ELC matches the minimum-norm selector in identification accuracy and more accurately ranks distractors by harm.

Most adversarial attacks on deep reinforcement learning (DRL) assume white-box access to the victim policy, which rarely holds in practice. This paper studies transfer-based black-box attacks on DRL: the attacker crafts observation perturbations on a white-box surrogate agent and feeds them to an unknown victim. We formulate the attack as return minimization under a per-step perturbation budget. We first show that transplanting transferable image-classification attacks (FGSM, MI-FGSM, and NI-FGSM) with a per-step objective yields perturbations that transfer but are no stronger than random noise of the same budget. We then propose a trajectory-level attack that optimizes a sequence of perturbations over a receding horizon through a differentiable model of the environment and a temperature-smoothed surrogate policy, with the same optimizers. On CartPole-v1 with ten DQN and DDQN agents and 100 surrogate--victim pairs, the trajectory-level attack outperforms per-step attacks and random noise in the white-box, cross-model, and cross-algorithm settings.

Gaussian-process limits are powerful in describing overparameterized machine learning models, yet their validity in structured tensor-network architectures remains unclear. Here we analytically present a moment-based approach that identifies precise conditions for the emergence and breakdown of Gaussian universality in tensor-network learning models, with a focus on matrix product states. We prove that in the large bond dimension limit, the learning models with both local and global observables converge to Gaussian processes, with explicit finite-size bounds on higher-order moment deviations. Whereas in the large physical dimension limit, the Gaussian universality no longer persists: while the models with local observables retain Gaussian-process behavior, those global cases exhibit persistent non-Gaussian corrections. Our results reveal that Gaussian-process behavior in tensor-network learning is controlled not only by parameter number, but also by architectural scaling, observable locality, and the spectral properties.

This paper asks whether VLA generalization is grounded in a global understanding of objects' physical properties that enables policies to adapt their motion to unseen setups, or if policies simply replay the motions they've learnt that happen to succeed in new setups. The former reflects genuine generalization; the latter reflects incidental robustness. We first examine awareness of physical properties in seven VLAs by applying linear probing and representational similarity analysis (RSA) to their activations. We find that physical properties, including mass, fragility, deformability, friction and size are less decodable than non-physical properties such as semantic category, material, sound and price in nearly every modality stream. Compared with their base VLMs, robot pre-training weakens the linear encoding of physical properties in the language stream. Neither pre-training nor downstream fine-tuning strengthens the alignment between physical-property differences and activation distances. We then ask whether the weak physical information present in these activations shapes the actions a VLA generates. In a controlled LIBERO case study, we increase the mass of an in-domain object and signal the change through language or vision. Most VLAs use similar lifting behaviour for the heavier and original-mass objects, leading to task success declines. The few exceptions change their behaviour in response to lexical or visual cues rather than to mass itself. These results suggest that VLAs encode physical properties weakly and do not reliably use them to adapt their motion.

Preparing quantum states from classical data can cost more than the computation they serve; most loaders tailor a circuit to each input. Here we show that the signals of a real dataset share structure that can be learned once and reused. Our quantum-native loader learns a low-dimensional description of a dataset and prepares every signal with one fixed circuit set by a few numbers. Across seven views of five public datasets it meets the targets of the strongest structured loader at equal gate cost with several times fewer numbers per signal. These numbers can be inferred from a random subset: in a preregistered blind replication the subset needed to come within ten per cent of full-signal accuracy stayed constant within a prespecified margin as signals grew sixteenfold, whereas the structured loader needed ever more. It declines what it cannot represent, covering fewer cases than that baseline and no electrocardiogram.

Normalized gradient descent is a widely studied adaptive optimization method. Most existing analyses focus on the best iterate or a weighted average of the iterates, whereas practical implementations typically return the last iterate. In this paper, we study the last-iterate convergence of normalized gradient descent for convex, $(ν,M_ν)$-Hölder-smooth objectives. For a constant stepsize, we establish an upper bound of $\mathcal{O}\bigl((\log^2(T)/T)^{(1+ν)/2}\bigr)$, which contains a logarithmic overhead relative to the known $\mathcal{O}\bigl(T^{-(1+ν)/2}\bigr)$ guarantees for the best and weighted-average iterates. For $ν= 0$, this overhead is known to be unavoidable. We complement this analysis with numerical results based on the performance estimation problem (PEP), investigating the finite-horizon worst-case behavior in the smooth setting and whether the logarithmic overhead reflects an intrinsic limitation of constant-step normalized gradient descent. We then show that a linearly decreasing stepsize yields a last-iterate guarantee of $\mathcal{O}\bigl(T^{-(1+ν)/2}\bigr)$, matching the order of the best-iterate/weighted-average guarantees without requiring knowledge of $ν$ and $M_ν$.

Recent work on multi-agent LLM systems reaches sharply different conclusions: some results show that a single agent with the same information and compute should dominate a delegated system, others that multi-agent gains grow with task depth. We argue that much of the disagreement comes from modelling different bottlenecks, and introduce a stylised reliability model built around two trade-offs. Decomposition reduces the burden of long contexts but incurs a handoff tax when information is compressed or transferred between agents. Redundancy gains from multiple samples, but its benefit depends on how much their failures are shared. With reasoning budget, verification, and task structure added, the model yields two crossover conditions: decomposition becomes preferable once the attention cost avoided by resetting context exceeds the handoff cost, and parallel sampling at equal budget is eventually preferable when its shared-failure floor lies below the error floor of one agent thinking longer. We connect these regimes to recent theoretical and empirical results. On a ledger-reconciliation task we measure the context-degradation curve and the handoff tax from single-agent and handoff runs alone. From these the model places the crossover at depth 10 and predicts decomposition to win at depths 20, 50, and 100. It does, on step-level and final-balance accuracy, and the decomposed system's success, which the prediction never sees, lands within 9 percentage points of the predicted rate at every depth.

Large Language Model (LLM) assistants routinely decide whether they can trust users and third parties whose competence, intentions, and integrity they cannot verify. This uncertainty matters for safety, as trusting the wrong party can lead an agent to comply with harmful requests or act on malicious instructions encountered during tool use. To study this problem, we define trust as an assistant's willingness to accept vulnerability to the actions of another party and ask whether such behavior can be causally controlled through model activations. We build 2,000 contrastive conversations spanning ability, benevolence, and integrity, where paired responses complete the same request but differ in whether the assistant trusts the user. From these pairs, we learn steering matrices while keeping the model parameters frozen and test them across six instruction-tuned models from three families, finding that steering changes trust decisions monotonically in both directions. We then ask whether this effect extends to several safety-related agent settings involving harmful requests, prompt injections, and insider threats, while using benign-task and reasoning as controls. Our findings provide evidence that trust in the user can be causally controlled along linear directions in model activations and provide a way to study how trust shapes safety-relevant behavior in language models.

Semantic segmentation of street panoramas can support detailed descriptions of urban environments, yet small datasets and unequal training costs make model selection difficult. This paper presents the system used for a first place submission to the PalmCity challenge in the leaderboard snapshot dated 5 October 2026. Nine pretrained segmentation systems are compared using approximately equal computation budgets. The candidates include DeepLabV3+, SegFormer, UPerNet, Mask2Former, DINOv3 with a linear decoder, and an Encoder only Mask Transformer using DINOv3. The two leading candidates are trained independently with three random seeds and longer budgets. Equal averaging of class probabilities from the three Encoder only Mask Transformer models, evaluated at three image scales with horizontal reflection, produces 60.95% mean intersection over union and 71.16% mean F1 on the 84 image public validation split. The submitted predictions receive 57.08% mean intersection over union and 67.96% mean F1 on the hidden test leaderboard. Producing all 249 test masks takes 251.49 seconds including model initialization and provenance checks on one NVIDIA RTX 5090. Peak allocated GPU memory is 2.70 GiB. The study reports all eligible models, all inference variants, class level errors, source conditions, and reproducibility checks, providing a documented challenge workflow with existing architectures.

Federated multi-player multi-armed bandit problems model collaborative sequential decision-making where multiple players interact with a common bandit environment and share information through a central server to accelerate learning. Existing federated bandit frameworks typically assume that all players willingly share their local observations with the server. However, this assumption is often unrealistic in practical settings where players are self-interested and may not participate in collaboration without explicit incentives. To address this challenge, we propose an incentive-aware federated bandit framework in which players receive rewards for sharing information with the server and incur costs when buying information from the server. We develop a UCB-based algorithm, termed Buying-UCB, that balances individual exploration and collaborative learning by incorporating both sharing incentives and information acquisition costs into the learning process. We theoretically analyze the proposed algorithm and derive upper bounds on the group regret and buying cost. Our analysis further characterizes the trade-off between fully collaborative federated learning and completely independent learning. Extensive numerical experiments validate the theoretical findings and demonstrate the effectiveness of the proposed framework under different collaboration and pricing regimes.

Modern text-to-speech (TTS) systems can clone a target speaker from a short reference clip or be fine-tuned on a target voice, yet their behaviour on morphologically rich, lower-resource languages such as Turkish remain under-characterised. We present a systematic benchmark of four contemporary systems (Chatterbox, CosyVoice, OmniVoice, and VoxCPM2) evaluated across fine-tuned and zero-shot configurations, contrasted with a conventional VITS baseline and anchored to natural gold speech. Each configuration is scored with eighteen complementary objective metrics spanning learned naturalness predictors (UTMOS v2, DNSMOS-Pro, SCOREQ, WhisQA, AudioBox-PQ, NatScore, SpeechLMScore), intelligibility and signal-quality estimators (SQUIM PESQ/SI-SDR/STOI, Brouhaha), speaker similarity, distributional fidelity (TTSDS) and low-level acoustic descriptors. We further analyse how quality varies with utterance length and quantify long-form temporal consistency through speaker-identity and naturalness drift over chunked utterances. We release our evaluation code to support reproducible TTS evaluation.

Large Vision-Language Models (LVLMs) frequently suffer from object hallucination. Existing training-free interventions primarily manipulate attention weights, which indirectly affect the deep semantics reaching the final predictive layers. In this work, we shift our focus to the hidden state vectors extracted after self-attention and residual addition. Empirical analysis reveals that hallucinated tokens do not simply over-rely on linguistic priors; instead, they exhibit an anomalous contextual deviation, showing significantly lower similarities to both textual and visual contexts in intermediate layers. Motivated by this, we propose ROT, a layer-specific, training-free framework. ROT dynamically detects semantic deviation in the middle layers and applies a norm-preserving rotation to steer the hidden states back toward the local multimodal context plane spanned by the contexts. For subsequent layers, a representational smoothing mechanism is introduced to stabilize the calibrated trajectory. Extensive experiments on multiple benchmarks demonstrate that ROT consistently reduces hallucinations across various model architectures and scales, offering an efficient, geometry-driven solution for grounded generation.

Learning-based mutant selection is used to reduce the cost of mutation testing by ranking and selecting the mutants that seem to be the most promising. Prior work has shown great potential for this approach in fault revelation and subsuming-mutant selection. However, the risks and benefits of these approaches -- especially the distribution of selected mutants across code locations and how that distribution affects behavioral diversity and fault revelation -- remain unclear. We investigate this distribution and its effects, and introduce LinePool, a simple, model-agnostic stratification step that distributes selections across source lines while retaining the underlying ranking signal. Across two datasets and budgets (2% and 5% of the mutants), we find that learning-based selection concentrates the budget on a few code locations, and may thereby systematically introduce 'blind spots', untested code areas, allowing faults to escape detection. On larger programs, the selected mutants also exhibit substantial overlap in kill behavior. LinePool substantially increases spatial coverage and reduces kill-set overlap. It also improves fault revelation at tight budgets and makes selection more robust to imperfect or shifted ranking signals. Comparisons with clustering and established diversification methods show that none consistently outperforms LinePool across datasets, supporting its use as a simple diversification step. In general, our results suggest that behavioral diversity should be considered alongside effectiveness when designing and evaluating mutant-selection methods, and that exploiting structural elements of the program is one simple way to obtain such diversity while maintaining effectiveness.

Three-dimensional (3D) left ventricular (LV) reconstruction from sparse cardiac magnetic resonance (CMR) imaging remains challenging due to inter-slice misalignment and insufficient local spatial information between slices. Global aggregation of contour features may obscure local contour-to-surface relationships. We propose Local2Mesh, a spatially localized contour-to-mesh framework that deforms a template mesh to reconstruct 3D LV geometry from sparse 2D contours without 3D mesh annotations. The framework introduces geometry-aware alignment to correct inter-slice misalignment and a plane-aware Local Router that routes contour features to template vertices using vertex-to-plane distances. Local and global contour features then jointly guide graph-based template deformation for 3D LV reconstruction. Experiments on two public datasets, M\&Ms-2 and ACDC, demonstrate superior geometric reconstruction and functional estimation over existing methods. Zero-shot transfer from M\&Ms-2 to ACDC demonstrates strong cross-dataset generalization. Reconstructed meshes also improve disease classification over sparse contours, supporting their utility for downstream cardiac analysis. These results demonstrate that combining geometry-aware alignment with local contour-to-vertex modeling improves LV reconstruction from sparse 2D contours and supports downstream cardiac analysis. The code is available at \url{https://github.com/hwu918945-alt/loca2mesh}.

Sparse autoencoders (SAEs) are increasingly used not only to interpret language models but also to intervene on their internal representations. We show that this creates a supply-chain attack surface: a maliciously modified SAE can induce attacker-chosen behavior when inserted into the forward pass of an otherwise unchanged language model. We introduce a decoder-only SAE backdoor that leaves both the underlying LLM and the SAE encoder frozen, restricting the attack to a single auxiliary component at a single insertion layer. Using code generation as a case study, we demonstrate high rates of unsolicited code insertion across three language models and a wide range of insertion layers, as well as trigger-dependent behavior conditioned on a prompt cue. We further evaluate the modified SAEs using HumanEval and selected SAEBench metrics. While attack effectiveness varies across models and layers, strong backdoor behavior can coexist with relatively small changes in several conventional SAE quality measures. These results establish that SAEs can carry behavioral backdoors without modifying the language model itself and should therefore be treated as security-sensitive components.

Agentic AI assistants increasingly act on developers' behalf by modifying files, executing commands, and accessing external resources. These actions often require permission, yet little is known about how practitioners make permission decisions while still benefiting from agent autonomy. To address this gap, we conducted a sequential mixed methods study, interviewing 18 practitioners who use AI agents and then surveying 115 practitioners based on the interview findings. We find that practitioners often understand agent behaviour through what they can directly observe and review, while decisions, data use, and other activity behind the scenes remain less clear. This uncertainty also shapes permission decisions, which depend on the scope and risk of an action, whether it fits the task, familiarity with the agent, and the environment in which it operates. Practitioners respond by adjusting how closely they oversee agents, from setting limits in advance to monitoring execution and reviewing work afterwards. How much scrutiny they apply depends on factors such as trust, task importance, time pressure, and the consequences of an action. Our findings suggest that permission systems should make consequential actions easier to review, distinguish what an agent is allowed to do from what the user intended, make reversibility clearer, avoid treating repeated approvals as stable preferences, and distinguish rejecting a single action from rejecting an entire approach.

Liquid-rocket thrust-chamber design involves interdependent analyses in which downstream constraints can require earlier design decisions to be revisited. Managing these dependencies across heterogeneous tools requires consistent design information and coordinated updates throughout the workflow. We present RocketAgent, a long-horizon engineering agent for multidisciplinary preliminary design of liquid-rocket thrust chambers. A single plan-owning Coding Agent coordinates engineering skills for performance sizing, subsystem optimization, geometry generation, and multiphysics assessment. A provenance-aware knowledge graph supports method selection, while a typed Design Intermediate Representation maintains shared parameters, artifacts, and decisions. Revision-aware checks invalidate affected results and block superseded inputs, with consequential changes subject to engineering approval. In a representative simulation-based design, RocketAgent continued from an infeasible cooling search through an engineer-authorized operating-point revision, identified feasible subsystem designs, and coordinated subsequent geometry generation and multiphysics assessment to support final configuration selection. Separate module tests assessed surrogate predictions and nozzle adaptation. A two-configuration comparison across three controlled scenarios verified the expected dependency invalidations and superseded-input blocking before solver execution. The representative case demonstrates sustained coordination across a multidisciplinary design workflow, while the controlled tests establish the behavior of the revision mechanisms supporting that execution.

Discovery of causal relationships in high-dimensional Gene Regulatory Networks (GRN) is computationally challenging and often difficult to interpret due to dense connections. Therefore, grouping genes together into functional modules can improve tractability and biological interpretability. However, existing cluster level causal discovery methods assume access to a predefined admissible partitions, requiring the graph over clusters to be acyclic. Constructing such partitions is therefore challenging. In this work, we introduce GO-based Clustering for Causal Discovery (GO4CD), an algorithm that uses Gene Ontology (GO) to construct biologically meaningful gene partitions at multiple levels of granularity, while favoring those more likely to be admissible for causal discovery. GO4CD groups together genes participating in a shared biological process, and propagates gene annotations through the ontology hierarchy to achieve different granularity of partitions. Furthermore, we integrate GO4CD with Causal Learning over Clusters (CLOC) algorithm and evaluate recovery of true Markov equivalence class both with an oracle of conditional independencies and on simulated gene expression data using multivariate conditional independence tests. We evaluate GO4CD on multiple E.coli regulatory subnetworks and find that it is inadmissible in 18.1% of the cases, compared with 65.3-82.3% for the semantic-similarity baselines. Our results indicate that GO4CD is substantially better suited to learning causal GRNs defined over biologically meaningful gene clusters.

Deep learning has advanced chest X-ray (CXR) diagnosis, yet demographic biases in learned representations may contribute to performance disparities across intersectional groups. We propose a single-encoder framework combining dual-level decorrelation with prototype-guided cross-group contrastive learning to reduce demographic dependence while accounting for within-class variation. We further propose Demographic Representation Alignment Reduction (DRAR), a new metric that quantifies the reduction in demographic structure within disease representations. The framework is evaluated on four classification tasks using 34,809 CheXpert test images across eight intersectional groups, defined by age, sex and ethnicity. Compared with empirical risk minimization (ERM), our method reduces the mean equalized-odds gap from 15.41\% to 10.86\% and the AUC gap from 5.95\% to 5.01\%. Our method achieves a DRAR of 59.04\% relative to ERM, with only a slight decrease in mean AUC. These results demonstrate that representation disentanglement can reduce demographic bias and improve intersectional fairness. Code is available at \url{https://github.com/06Yujie/Fair-Medical-Imaging}.

Recent work has shown that probabilistic flow matching for time series forecasting benefits from a data-matched prior. The resulting prior introduces local correlations, which a sequential architecture usually absorbs: a recurrent neural network (RNN), a structured state-space model (S4), or a Transformer. However, such a backbone costs GPU memory and time per epoch. A cheaper alternative is MLP-based latent-space flow matching: embed the time series via an invertible map to a single latent vector and learn the flow there, so a tabular MLP can treat the series as a set of features. The relationship between the prior and the choice of linear latent map is understudied in conditional flow matching (CFM) forecasting, yet we found it strongly affects performance. Fixed transforms such as Fourier or discrete cosine (DCT) are only well-conditioned for Ornstein--Uhlenbeck priors, while a principal-component (PCA) map fit to the data is a strong but training-set-dependent reference sensitive to train--test shift. Instead, we propose to use the Mercer eigenbasis of the prior kernel: it diagonalises the centred covariance exactly, decouples from training data, and adapts to non-stationary and periodic priors. On five GluonTS benchmarks (ETTh1, ETTh2, Weather, Electricity, Traffic) under a shared protocol with TSFlow, the resulting MLP matches or beats it on CRPS at about $4.7\times$ less training memory and $3.5\times$--$4.4\times$ less time per epoch.

We consider the fusion of lung ultrasound images with routinely-collected clinical and demographic data for the purpose of automated tuberculosis (TB) screening using deep-learning. Such deep-learning based screening tools for TB could meaningfully support the health care system in Africa, where the burden of disease is severe and resources are constrained. Beginning with an established ResNet baseline for classification of lung ultrasound images, which achieves an area under the receiver operating characteristic (AUROC) curve of 0.91 [0.86,0.96] (95% CI), we consider the incorporation of the clinical and demographic data using three fusion approaches. We find that a simple average-based fusion of the output scores of separately-trained image and clinical data classifiers consistently matches or outperforms a more complex approach where the data is fused earlier and a combined classifier is trained. Fusing the image and the clinical classifiers in this way leads to a classifier with an overall AUROC of 0.95 [0.91,0.99] (specificity of 0.76 at sensitivity 0.93) which is an improvement of 4% absolute over the image-only baseline. We also find that greedy feature selection can be used to reduce the number of clinical and demographic inputs without sacrificing classification performance. Finally, when we differentiate between clinical and demographic data that are self-reported, that require some basic measurement or calculation, and that require a point-of-care (POC) test, we find the inclusion of the POC tests included in this study to be of minimal benefit to classification performance. We conclude that the incorporation of routinely-collected clinical and demographic data is a promising way to improve the performance of lung ultrasound based automatic classification.

Physics simulators can support the discovery of quantum experiments by predicting the states generated by experimental configurations. When these simulators are computationally expensive, repeated simulator calls can limit the search for experiments that generate a desired quantum state. Here, we develop a physics-inspired polynomial neural surrogate for PyTheus, a graph-based quantum-optics simulator, to predict quantum states and use it to design quantum experiments. Its polynomial activations are motivated by the relation between graph perfect matchings and the resulting state amplitudes. We train separate surrogate models for four-, six-, and eight-photon systems and show that they achieve higher prediction accuracy with fewer trainable parameters than standard multilayer perceptrons. We then use the trained surrogates for inverse design of GHZ, W, and linear-cluster states. For the larger systems, the surrogates also enable faster inverse design than direct optimization with PyTheus. These results suggest that incorporating the underlying physics into neural surrogates can provide an efficient approach to quantum experiment design.

Next-token prediction (NTP) is the standard pretraining objective for large language models, yet it provides an explicit training signal only for the immediate next token, which can lead models to exploit local patterns instead of capturing longer-range structure and ideas. Multi-token prediction (MTP) addresses this by training models to predict several future tokens. However, existing MTP methods often introduce a large number of new parameters with limited improvements in downstream performance. Latent MTP approaches address this efficiency issue by encoding future tokens into a vector representation. However, these approaches usually rely on external helper models for future token encoding. We propose LightMTP, a lightweight, i.e., parameter-efficient, latent MTP approach that bootstraps the future token representations from the model's own hidden states. Our two LightMTP variants extend supervision to more future tokens without requiring the additional computational overhead of conventional MTP nor the external supervision latent MTP normally relies on. LightMTP adds at most 1% extra parameters, retains better performance on general language modeling benchmarks, and achieves similar gains in planning, coding, and reasoning.

SVD-based pruning and quantization have recently emerged as a promising strategy for the ultra-efficient compression of large language models. In these methods, compression is performed in two stages: components are first truncated, and the remaining ones are subsequently quantized. Although this decoupled pipeline benefits from both pruning and quantization, it requires separate optimization for each stage and fails to fully exploit their balance, which can lead to suboptimal performance under aggressive compression. To address this limitation, we propose a new LLM compression method that co-optimizes pruning and quantization in a unified framework. Our key idea is a differentiable method for learning component-wise bit-widths, allowing less important components to be assigned 0-bit precision and pruned away. Notably, our method performs favorably against two-stage baselines, even when subjected to extreme quantization settings ($1.61$ bits) designed for ultra-efficiency. Code: https://github.com/MMAI-Laboratory/DBW.

Joint attention - the human ability to share a common visual or cognitive focus with others - enables a meeting of minds that lets us coordinate even with unfamiliar partners. In this work we investigate whether equipping AI agents with a similar mechanism can enable such zero-shot coordination. We introduce Mutual Attention for zero-shot TEaming (MATE): a novel multi-agent reinforcement learning method inspired by human joint attention. MATE encourages agents to coordinate their actions by aligning their visual attention on scene-salient objects during the interaction rather than relying on arbitrary partner-dependent conventions established during training. Unlike symmetry-breaking approaches that merely prevent brittle conventions from emerging, MATE actively promotes coordination through an environment-grounded signal that is naturally shared across partners. We evaluate MATE on three benchmarks: our Card Alignment Game, designed to isolate brittle convention formation, and the more challenging Level-Based Foraging and OvercookedV2 benchmarks. Our experiments consistently show that a joint-attention-inspired signal improves coordination with unknown partners, underlining MATE's potential as a general coordination mechanism that complements and surpasses symmetry-breaking approaches.

Exploiting meaningful latent structures from data to solve downstream tasks is a fundamental challenge in signal processing and machine learning. While Principal Component Analysis (PCA) and coVariance Neural Networks (VNNs) successfully leverage the covariance matrix to process data, they inherently capture both direct and indirect correlations. The precision matrix (inverse covariance) overcomes this by explicitly encoding conditional independencies, making it largely studied in graphical lasso and graph topology identification. However, finite-sample precision estimates are notoriously unstable, and regularized estimators remain task-agnostic. In this work, our principal contribution is tackling the challenging problem of task-aware graph inference. We propose Precision Neural Networks-Joint (PNN-Joint), a framework that jointly estimates a sparse, statistically grounded precision matrix alongside graph neural network weights via an alternating optimization scheme. As a foundational framework to support this, we introduce Precision Neural Networks (PNNs), a broader class of graph convolutional networks operating on precision estimators, and establish their spectral connections to PCA and VNNs alongside their stability to finite-sample errors. Extensive empirical evaluations on synthetic data, as well as real-world neuroimaging and motion sensor datasets, demonstrate that PNN-Joint yields highly interpretable task-aware graphs, exhibits remarkable robustness in low-data regimes, and consistently achieves the best or second-best performance among competitors on real-world tasks.

Image editing should change only the attributes specified by an instruction while preserving everything else, yet current methods often make unintended changes. We treat this minimal-change principle as an optimization objective for instruction-based editing. Latent L1 regularization is a poor proxy for output locality in modern nonlinear generators and often requires supervision unavailable at scale. We instead optimize edit outcomes with reinforcement learning. An agentic vision-language reward model audits each source image, instruction, and edited image for two failure types: unimplemented requested changes and unintended changes. A group-level rubric merges and verifies these issues to provide consistent rewards across candidate edits without per-instruction human annotations. On FLUX.1 Kontext-dev, ARRO raises average EditScore from 5.21 to 5.88 across MinEval, MagicBrush, AnyBench, and Emu-Edit. On 600 evaluation examples, it reduces off-target pixel change by 8.4% relative to the base editor. Reward and SFT controls, blinded human evaluations, and transfer to OmniGen2 provide complementary evidence. Code: https://github.com/Showwwwwwwww/ARRO

With massively parallel simulation, on-policy Reinforcement Learning methods such as PPO have become standard in many domains. However, learning from scratch is sample-inefficient and fails to exploit the potential existence of a suboptimal expert, such as a heuristic, a model-based controller, or a policy trained on a related task. Such an expert is often available and can guide early training, but its sub-optimality limits final performance. The challenge then becomes balancing expert guidance against learning from rewards. Existing methods set the expert's influence through a blending weight, a schedule, or an evaluation-driven curriculum. Alternatively, they adapt it with additional learned components such as critics over expert actions or auxiliary agents. However, none optimizes it using the same on-policy objective as the policy itself. We propose a method in which the learner and the expert alternate control within each training episode, and the expert's share of control is a single learnable parameter optimized jointly with the policy. The learner benefits from the expert early in training, but its share of control declines as the learner becomes more competent, until eventually vanishing completely. This leaves the learner acting alone and better than the suboptimal expert. We evaluate our method on 34 tasks across two benchmarks, spanning discrete and continuous action spaces, using both learned and model-based experts. Our method improves sample efficiency over guided and unguided baselines while requiring minimal hyperparameter variation. The expert's share decays to zero as the learner improves, vanishing when the expert is no longer useful.

Spatio-temporal video grounding (STVG) aims to localize objects or events described by natural language queries in both space and time. Existing STVG models are typically trained and evaluated under the assumption that each query is relevant to the input video. In this work, we challenge this assumption by studying the behavior of state-of-the-art STVG models under irrelevant queries and missing textual input. Our experiments show that current models can still produce plausible spatio-temporal predictions even when the query is unrelated to the video or removed entirely. We further analyze HCSTVG-v2 and VidSTG to identify dataset regularities that may encourage such query-insensitive behavior. Our study highlights an underexplored limitation of STVG models and motivates negative-aware evaluation protocols and architectures that explicitly assess query relevance.

Agentic buyers can cheaply fork a procurement task into many parallel negotiations, but concurrency is not free: every thread consumes resources, and simultaneous agreements create cancellation and commitment risk. We study a one-unit post-order sourcing problem with a single hard-deadline negotiation window, in which a planner jointly chooses the number of seller-facing negotiators and a common procurement price cap. The model combines a product-specific acceptance curve with fulfillment loss, per-thread cost, and excess-commitment cost. We establish three structural results. First, holding the per-thread acceptance target fixed, the marginal value of another negotiator decays geometrically, yielding a conditional concurrency threshold. Second, under a convex quantile curve, parallelism substitutes for concession: more concurrent negotiators imply a weakly lower per-thread acceptance target and price cap. Third, when prices are more dispersed, Agentic buyers benefit by searching harder for bargains, but suffer when they instead try to guarantee procurement by offering higher prices. We operationalize these results in the Concurrency-Aware Negotiation Optimizer (CANO), a deterministic optimizer that jointly determines the optimal negotiation concurrency and procurement price cap for an agentic procurement system. Across different analytic market configurations and extensive Monte Carlo, finite-data, non-Gaussian, and correlated-seller stress tests, CANO consistently outperforms common heuristic policies while validating the predicted structural properties.

In this work, we study transformer attention through the lens of spectral geometry and operator theory. We view each attention head as a functional map between Hilbert spaces of functions on the token sequence and derive a Token Difference Operator, whose spectral structure controls how token-space information is routed to the output. We show that standard Euclidean spectra are structurally biased by sinks, conflating mass concentration with genuine routing capacity. By recasting token space in the intrinsic probability geometry induced by attention, the token difference spectrum disentangles sink effects from routing capacity and provides a spectral description of the dimensionality of the head output. This yields a unified framework for analyzing attention maps, explaining sinks, routing collapse, and output dimensionality within a single operator-theoretic framework. In practice, by grounding attention heuristics in spectral geometry, we develop a novel attention-based uncertainty estimator that complements probability-based scores, with the largest gains on long-context inputs.

Block diffusion accelerates speculative decoding by drafting multiple tokens in one forward pass. However, each position predicts a marginal distribution without observing earlier proposed tokens, limiting draft quality and acceptance length. We identify a concrete failure, the \emph{repetition trap}, in which neighboring positions produce redundant copies of the same token. We explain this tendency theoretically and empirically examine its association with shorter accepted drafts. Recent methods refine marginal predictions with an additional causal head or a separately trained drafter, increasing parameter storage and introducing separate training objectives. We instead propose D-Loop, which introduces \emph{intra-block causal conditioning} within the original diffusion drafter without additional model components. Inspired by semi-autoregressive generation and parameter sharing, D-Loop reuses the same backbone across looped passes. The first pass proposes a block, and the second conditions on a selected prefix to regenerate the suffix in parallel. A complementary prefix--suffix objective trains the shared drafter for both anchor-only prefix prediction and prefix-conditioned suffix prediction. Across eight math, code, and chat benchmarks, D-Loop can beat DFlash and DSpark on Qwen3-4B and Qwen3-8B with obvious gains.

Ultrasound is widely used, but acquisition quality is heavily dependent on the operator's knowledge and expertise. With demand for examinations outpacing the supply of trained sonographers, operator-guidance systems aim to close this gap by instructing a less trained user how to move the probe toward a target view. In this paper, we propose a retrieval-induced latent transition model for ultrasound acquisition dynamics, formulating ultrasound operator guidance as multi-step planning and retrieval in a world model. Using a V-JEPA 2.1 backbone, observations are first encoded into a latent space where anatomically related views lie close together. We then retrieve similar views from a reference database containing encoded latent states and corresponding probe positions and orientations. Rather than learning a parametric transition function, we directly use physically executed transitions from the database to establish our nonparametric, retrieval-induced transition model that supports receding-horizon planning. At deployment, guidance is generated from the live ultrasound image feed alone, without any probe tracking hardware. Applied to carotid ultrasound, the proposed planner reaches the target view in 86% of retrospective closed-loop episodes, versus 52% and 43% for representative baselines, outperforming both on every target view, including the challenging longitudinal internal and external carotid artery views. A prospective feasibility study on unseen volunteers, run in real time on a CPU using distillation, reaches 83% target-view reachability. Because planning is driven by proximity to any encodable goal latent, the same world model can navigate back to any previously acquired, patient-specific frame, supporting reproducible longitudinal imaging for e.g. perioperative or follow-up monitoring.

We study contextual queueing bandits, where a learner schedules jobs while learning unknown service rates modeled by logistic functions of job-server features. Performance is measured by queue length regret, the expected excess queue length at round $t$ relative to an oracle that knows the service rates. Existing decaying-regret guarantees either have a suboptimal decay rate or require a known fixed horizon. They also assume context-wise slack and a strictly positive minimum eigenvalue of the feature covariance. In this paper, we propose WISE (Widest Interval Selection with Elimination), achieving rate-optimal $\widetilde{\mathcal O}(t^{-1/2})$ queue length regret at every sufficiently large time without knowing the horizon. We assume capacity slack, meaning that expected incoming workload under best-server service is below service capacity, and impose no covariance lower bound. Our analysis uses a workload potential measuring the expected service attempts needed by waiting jobs on their best servers. Its drift on nonempty rounds combines a negative term ensured by capacity slack with errors from suboptimal service choices. Then an elliptical potential count bounds how often WISE selects wide confidence intervals, thereby limiting the number of rounds with large service errors. We also sharpen the arrival-rate dependence of an existing lower bound and make its dependence on feature dimension and server count explicit. We prove another lower bound that quantifies the increase in regret as the normalized capacity slack decreases; to our knowledge, this is the first such lower bound for CQB. Simulations show small regret even when context-wise slack fails.

Small independent (indie) game startups face fierce competition when developing games with limited time, budget, and resources. Continuously testing and experimenting with new ideas, features, and content is essential to validate and guide development toward a viable and successful game. Such continuous experimentation (CE) requires access to users and user data, which is often very limited in the early stages of development, both in general and, in particular for indie game companies. We provide insights and recommendations on how to perform experimentation and game testing in pre-release stages of indie game development based on an exploratory interview survey with 10 independent game developers from different companies. We provide a framework for CE that highlights key parts to consider when planning and implementing an experiment. Current practices for experimentation within indie game development are also described for the early development stages of ideation, preproduction, feature production, and content production. Our empirical findings offer insights and recommendations into how to conduct experiments during pre-release stages of game development, specifically for Indie game companies. Other software startups, particularly those with a value proposition centred around interactive and satisfying user experiences, can also benefit from our research.

Diffusion models provide expressive priors over trajectories, but adapting these priors to test-time constraints requires maintaining feasibility and consistency across successive control decisions. We introduce Persistent Particle Planning (P3), a sequential Monte Carlo framework for diffusion control that maintains a weighted population of candidate plans across replanning steps. At each control step, P3 shifts and partially re-noises the candidate trajectories, refines them under the latest observation, and uses constraint-aware weighting and resampling to select among alternative continuations without retraining the diffusion model. We consider denoising and replanning as one Feynman--Kac particle system and analyze it under an idealized repair. We prove that the re-noising depth controls how reliably a kept plan stays on its route, and that keeping a rare, well-separated route takes far fewer plans than rediscovering it by sampling from scratch. Experiments under multiple test-time constraint configurations show that population reuse reduces route switching and improves success without constraint violations. Because P3 refines earlier plans instead of redrawing them, it also needs fewer denoising iterations per replan. On maze-navigation tasks, it plans faster than both regenerated populations and methods that correct a single sampled plan by constrained optimization. Code and pretrained models are available at https://github.com/p3-username/p3-anon.

AI agents built on large language models (LLMs) run shell commands, read and write files, and reach the network, typically with their user's privileges. However, what an agent does during an execution is difficult to understand: tests assert on the result, and the agent's trajectory records only what the agent reports about itself, which may omit behavior executed by its subprocesses. We present AgentSpy, an approach that observes an agent from outside the agent. AgentSpy runs the agent in an isolated environment, configured by a declarative specification, and records the system calls and network traffic of the agent and of every process it executes. Based on this monitoring, AgentSpy supports two families of analyses: conformance analyses, which measure obligations, i.e., what an agent execution should do, and safety analyses, which check prohibitions, i.e., what an agent execution must never do. We instantiate one analysis of each family. The reliability analysis uses rules to summarize each run by the environment resources the agent uses: the commands it executed, the files it accessed, and the hosts it contacted. The security analysis applies deterministic rules to the system calls of an execution. For reliability, we evaluated AgentSpy on 77 tasks with the codex harness and three recent LLMs, executing each task three times. Sets of repeated runs of the same task are more similar than sets that include runs of another task in 92.2% of the comparisons. Among tasks for which all three runs pass outcome-based tests, the agent performs task-unrelated activities in 18% of the cases, reads the grading files in 7%, and does not use the developers' guidance in 17%. For security, the generic rules of AgentSpy detect four of five attack categories we considered, with no false positives across 50 runs.

While AArch64 CPUs are becoming strong market contenders, their software ecosystem lags behind the mature x86-64 environment, hindering the adoption of the new architectures and impacting user experience. Binary translation bridges this divide by converting binary code from one architecture (e.g., x86-64) to run on another (e.g., AArch64), allowing legacy software to benefit from modern hardware's performance and energy efficiency advantages. Current translation methods are typically either dynamic, which adds significant runtime overhead, or static, which struggles with reliability due to the inherent complexities of binary analysis. This paper introduces a new static, assembly-to-assembly translation paradigm that transforms binary code ahead of execution, generating portable, efficient native-like binaries that run on AArch64 devices without runtime frameworks. Benefiting from recent breakthroughs in large language models (LLMs), we provide a practical and automated translation engine that produces high-quality code with minimal human intervention. To ensure correctness, we introduce a crucial verification step, where we split the assembly code into simplified snippets, enabling efficient and scalable semantic verification. Our evaluation shows that this approach significantly outperforms existing open-source solutions with a large margin, producing binaries with near-native performance. Furthermore, it shows substantial improvements over the leading industrial translator, ExaGear, illuminating a promising new direction for cross-architecture binary translation research.

Molecular dynamics simulations proceed by integrating the Langevin equations over many small femtosecond timesteps. This poses a challenge for estimating ensemble properties and transition dynamics that occur on much longer timescales. We introduce Langevin Flow Maps, which extend machine-learned force-fields to additionally learn the stochastic Langevin integrator. We show that Langevin Flow Maps enable large-timestep molecular dynamics and recover accurate dynamical properties of the system, while running an order of magnitude faster than current machine-learned force fields. Further, by training on a diverse molecular dataset, we demonstrate a path towards transferable Langevin Flow Maps.

Latent world models based on Joint-Embedding Predictive Architecture (JEPA) are deterministic by design. While successful in fully observable scenarios, this paradigm breaks down when past observations and actions lead to multiple plausible future possibilities, e.g., due to occlusion. We introduce EpicWorldModel, a framework to train stochastic JEPAs for environments and tasks with inherent uncertainty under partially observability. We jointly train the EpicWorldModel predictor with its latent representation space to directly predict multiple potential future states using a flow-matching objective, when the goal-relevant scene content is absent from the conditioning history. We show that flow predictive variance, motivated by its relation to an upper bound on predictive entropy, serves as a useful exploration guidance for planning. By incorporating this uncertainty signal into Cross-Entropy Method (CEM)-based planning, our approach balances goal-reaching with exploration of uncertain regions where occluded goals are most likely to be located. We demonstrate the effectiveness of EpicWorldModel through a series of latent planning experiments with the best or on-par performance across tasks, showing up to 22% empirical improvement in success rate over LeWorldModel.

Vision-language-action (VLA) models currently demonstrate strong performance in a wide range of real-world robotics tasks. However, they often still lack the generalization ability to handle large visual out-of-distribution shifts. Post-training of VLAs with reinforcement learning (RL) has been shown to benefit robustness, but significant room for improvement remains. In this work, we systematically study the effect of image augmentation on VLA post-training. We find that it is crucial to augment only the critic module during RL updates, while leaving the actor's input clean during both rollouts and updates. For $π_{0.5}$ and GR00T N1.5 this raises out-of-distribution success on LIBERO-Plus by $7.8$ and $10.0$ points respectively, while augmenting the actor collapses training entirely. We investigate a range of augmentation types and strengths, and provide practical recommendations for improving generalization in VLA post-training.

Trust is crucial in the digital world, particularly in the age of systems based on artificial intelligence. Traditionally, system correctness is demonstrated by proving that the system satisfies the required properties. However, digital systems involve humans, who can make mistakes, break rules, or even intentionally cheat the system. Technical devices may also malfunction or cause errors. Therefore, external stakeholders need to trust a system throughout its operation. To address this challenge, we propose the concept of continuous trust. In this paper, we present a detailed explanation of this concept, based on a running example of a business's refund department. The paper concludes with a discussion of related work, such as compliance checking and certification, as well as an outlook on further questions regarding the formalization and the necessary tool support.

With the rapid development of artificial intelligence, the emergence of various Large Language Models (LLMs) has created a rich model ecosystem. However, this also brings a key challenge: how to select the optimal model for a specific user query. LLM routing addresses this need by dynamically assigning queries to the most suitable expert in the pool of candidate models. However, existing routing frameworks often simplify this process to a standard classification task; thus, a critical vulnerability is exposed when multiple candidate models correctly answer the same query. We formalize this capability overlap as routing noise, which misleads the router with arbitrarily correct candidate models, ultimately leading to routing collapse (a severe decline in generalization ability on unseen tasks). To address this problem, we propose a novel Cluster-Aware Soft-Labeling Routing (CASLR) framework. CASLR shifts the evaluation paradigm from the success of a single query to macro-domain consensus by replacing traditional one-hot vectors with a masked softmax mechanism. Specifically, for experts who answer incorrectly, we penalize their target probability to zero; for the remaining candidates, we directly compute continuous fine-grained soft labels based on their global clustering utility scores. We then use these refined soft labels to supervise a lightweight router. Specifically, the framework not only demonstrates superior accuracy on multiple benchmarks, but also outperforms Llama-3.3-70B-Instruct by 7.80% in overall average performance. Furthermore, the extremely low routing inference latency of only 1.13s further confirms that CASLR can achieve efficient system scheduling with almost zero additional overhead, while ensuring high response quality.

A prior longitudinal study of silent failures in a production LLM agent runtime (arXiv:2606.14589) found that about 70% were discovered by a human looking at the product as a user while thousands of tests and governance checks stayed green, and posed mechanizing part of what the human eye does as an open problem. This paper reports our attempt. We built an automated user-viewpoint observer targeting the most dangerous class, fail-plausible failure, in which an internal error becomes fluent, plausible output to the user. It is a two-layer pipeline: five deterministic signals distilled from incident postmortems escalate to an LLM judge whose verdicts must cite verbatim evidence or be discarded. Ground truth is 24 labeled production postmortems with explicit honesty boundaries: 16 of 24 are structurally invisible to any content-reading observer, and we say so. Offline, the deterministic layer achieves 6/6 regression detection with 0/4 false positives, each detector proven load-bearing by sabotage; held-out recall on novel patterns is 0/4. It is, so far, a regression engine. Deployment followed pre-registration: shadow mode caught and retired one systematic false positive on its first run, the 26-day shadow window then ran clean, flip criteria were fixed before the shadow data was read, and the analysis protocol was frozen before the enforcing-mode window opened. That window (12 observed days) fired zero verdicts; per the pre-registered path we report live precision as undefined rather than narrating quiet as success. The observer also produced ten silent failures of its own, confirming that the judge inherits the taxonomy it judges. We release the corpus, detector, and scorecard as a runnable bench: mechanization today retires the human's regression scanning so the eye can specialize in novelty; prediction remains open but is now measurable under frozen rules.

Tokenizer-free language models remove the inductive bias of fixed tokenizers by modeling text directly as bytes, but the resulting longer sequences substantially increase computation and eliminate explicit text abstractions. We ask whether this additional computation can be useful, and whether standard Transformers can learn the abstractions that tokenization provides. We study these questions on Transformers without specialized tokenization-related architectures. With token-superposition training and hash embeddings, byte Transformers consistently outperform subword Transformers as model size scales. We further find that byte Transformers build local text abstractions as external tokenizers: a set of segmentation-like positions are used to collect local context representations, and restricting up to $25\%$ of intermediate layers to these local representations preserves downstream performance. Finally, these learned structures induce highly non-uniform generation difficulty, with uncertainty concentrated near local structure boundaries; exploiting them for speculative decoding yields $3.4\times$ more accepted tokens than in subword Transformers.

Serving a large language model (LLM) across a fleet of deployments requires several weight-precision operating points. Multi-precision formats serve them all from one stream whose prefixes are valid lower-precision codes, instead of storing multiple copies. We present StagQ, a multi-precision weight format whose main stream is a 2-bit group-wise affine base followed by a configurable number of 1-bit refinement planes on a dyadic step schedule. Every supported precision is a readable prefix, decoded by an affine map derived from metadata shared across all precisions, with no per-weight lookup. A sparse side record, filled both before and after the grid is fitted, holds out the few weights the grid serves worst. We report two configurations of the encoder. At two bits the cheaper one leads the strongest multi-precision baseline on Llama-3.1-8B, Phi-4, and OLMo-2-7B by 3.1 to 7.0 MMLU points, at a slightly lower logical rate. At three bits it leads on Llama-3.1-8B, leads on Phi-4 at a higher rate, and ties on OLMo-2-7B. At four bits it ties on all three, at a higher rate. In a batch-one matrix-vector product on an NVIDIA A100 GPU, timed on synthetic weights, our kernel is faster than the two baseline kernels in most shape-precision cases.

Verifier-guided decoding can prevent harmful reasoning steps from contaminating subsequent generation, but typically relies on an external learned verifier. We ask whether a language model can instead reject its own bad reasoning steps. We define a prefix's recoverability as the probability that the frozen generator can complete it correctly. Diagnostics show that adjacent recoverability changes are often difficult to resolve with practical Monte Carlo budgets, while same-prefix candidates exhibit a sparse low-recoverability tail. We introduce Self-Step Rejection (SSR), which trains a lightweight LoRA acceptance gate on the generator backbone while keeping the base model frozen. SSR uses confidence-qualified first-passage supervision: steps before the first resolved crossing of a root-relative recoverability barrier are accepted, the crossing step is rejected, and unresolved steps and suffixes are excluded. Training combines pointwise classification, same-prefix pairwise learning, and group-relative policy refinement using final-answer correctness. At inference, SSR accepts candidates or resamples from the unchanged prefix under rejection budgets, without an external learned verifier. Across three reasoning models and five mathematical reasoning benchmarks, SSR improves macro-average accuracy over single-pass decoding by 5.4--10.1 points using 1.21--1.40x as many generated tokens, and achieves the highest macro-average accuracy among evaluated step-level methods. Full-solution scaling methods require 4.47--8.27x the single-pass token cost for comparable performance.

Reinforcement learning can substantially improve a reasoning teacher, but it is unclear which of those improvements survive when the teacher supervises a smaller on-policy student. We study this question in mathematical reasoning by comparing teacher lineages before and after GRPO, multiple student scales, direct GRPO, and several on-policy distillation objectives. The central finding is that transfer is structured rather than scalar: teacher strength alone does not make dense distillation competitive, while an RL-improved teacher creates useful but metric-dependent student gains. This motivates Transfer-Stratified On-Policy Distillation (TS-OPD), which screens training problems by the joint sampled success of the student and teacher, routes acquisition problems to gated forward KL, routes consolidation problems to gated reverse KL, and adds an entropy brake to protect sampled coverage. Across the main comparison, TS-OPD is the strongest student objective for macro average correctness with the GRPO-improved teacher, while pass@K remains more mixed. Ablations show that the gains come from routing and token gating rather than skipping problems. These results support a transfer-aware view of OPD: stronger teachers help when the supervision direction and token budget match the student's observed ability, not merely because the teacher endpoint is stronger.

Diffusion policies provide expressive action distributions for continuous-control reinforcement learning. However, safety-aware online diffusion policy optimization remains underexplored, particularly methods that use predictive reachability information without an explicit dynamics model. We propose Reachability-Aware Diffusion Policy Optimization (RADPO), a model-free method that combines predictive first-hit safety estimation with cumulative-cost budget feedback. RADPO learns a discounted first-hit reachability value that captures the discounted risk of a cost event, assigns larger weight to events that occur sooner, and uses this signal to shape the reward. A separate dual-like multiplier adjusts the shaping strength according to realized episodic costs relative to a prescribed budget. The diffusion actor improves through weighted denoising regression on candidate actions scored by the reward critic. Our approach requires neither a learned dynamics model, action gradients through the critics, nor differentiation through the reverse diffusion sampler. We establish theoretical properties of the reachability value and show that accumulated reachability penalty provides a conservative surrogate for future discounted cumulative cost. Across ten continuous-control safety tasks, RADPO achieves competitive reward-cost trade-offs, with substantial reductions in constraint violations on several tasks relative to the compared baselines. Our theoretical and empirical analysis supports that combining reachability with cumulative budget feedback is a viable approach to safety-aware diffusion policies.

We study last-iterate convergence in unknown two-player zero-sum discounted Markov games with bandit feedback. The players learn independently along a single trajectory without observing each other's actions. We develop Adaptive Regularized TD Learning (ARTD), which achieves a $\widetilde{\mathcal{O}}(t^{-1/4})$ duality gap bound for the current policies under a uniform hitting time assumption, with high probability simultaneously over all rounds and starting states. This improves the $\widetilde{\mathcal{O}}(t^{-1/(9+ν)})$ rate of Cai et al. (2023), for any fixed $ν>0$, under the same feedback model and hitting time assumption. Our algorithm requires no knowledge of the hitting time bound, the time horizon, or the confidence level. To stabilize policy learning as value estimates change, we separate fast temporal difference averaging from bounded value updates. We adapt log-barrier regularization to the progress of value estimation, controlling both policy and value errors throughout learning. Together, these mechanisms enable fast convergence of the policies actually played, even when the players learn independently from bandit feedback.

Image compression, restoration, and generation all require a way to measure how different two images look to a person. Pixel error ignores how people see, while the most accurate perceptual distances are typically fitted to human judgments, tying them to a fixed data and resolution. For example, when image resolution is doubled, the correlation of DISTS with human scores on TID2013 drops from 0.815 to 0.717. We introduce the Jacobian Lens Distance (JLD), which derives its perceptual geometry from a frozen vision encoder rather than from human labels. JLD combines the locality of early patch features with the perceptual sensitivity captured by later encoder representations. Specifically, we use the encoder Jacobian to identify directions in the early feature space that most strongly affect the encoder output, producing a fixed metric tensor, $E[J^\top J]$, which we call the Jacobian lens. The lens is fitted only once from 100 unlabeled images, taking about 35 seconds. Locally, this construction defines a pullback metric in pixel space, giving JLD a clear geometric interpretation that can be directly analyzed on real images. Across four standard perceptual databases, JLD achieves state-of-the-art performance and consistently outperforms LPIPS, DISTS, PieAPP, and DreamSim. JLD is also robust to changes in image resolution, on TID2013, its lens-term correlation remains nearly unchanged when the resolution is doubled, decreasing only from 0.850 to 0.845. We further introduce JLD-fast, which is $4\times$ faster than LPIPS-VGG while achieving a mean correlation of 0.911. Finally, JLD naturally extends to video, reaching a correlation of 0.786 on Waterloo IVC 4K compared with 0.611 for VMAF.

Domain adaptation aims to turn a general-purpose large language model (LLM) into an expert for a target domain. While the dominant SFT+RL pipeline offers a convenient cold start, it may reduce exploration diversity and introduces additional complexity through multi-stage optimization. These limitations motivate RL-only adaptation. However, pure on-policy RL suffers from a cold-start problem, while mixed-policy RL still falls short: informative tokens in teacher outputs are learned too slowly in early training, and stale teacher outputs can hinder later improvement. We identify these two failure modes as Gradient Starvation and Teacher-Distribution Anchoring. To address them, we propose One-stage Policy Optimization (OnePO), which treats teacher outputs as transient guidance for policy improvement. OnePO combines Adaptive Objective Evolution to strengthen learning on informative low-probability teacher tokens and Teacher Retirement to discard teacher outputs once the current policy can surpass them. On medical adaptation, OnePO achieves 67.2 on HealthBench (Total) with only 20K training samples, outperforming SFT+RL and pure RL by 2.7 and 7.4 points, respectively. We further scale OnePO to produce HuatuoGPT-3, an open-source medical LLM series whose 27B variant reaches 70.1 on HealthBench (Total) and 71.4 on HealthBench Professional, surpassing frontier models such as GPT-6 Astra. Models and code are available at https://github.com/FreedomIntelligence/HuatuoGPT-3.

Valuing player actions in football requires accounting for strategic interactions among 22 players, including off-ball movements and defensive positioning. Existing reinforcement-learning-based methods commonly aggregate decisions at the team level or estimate player values independently, leaving strategic interdependence among players insufficiently represented. This study proposes an action valuation framework inspired by Markov perfect equilibrium (MPE) for all players. Each possession is modeled as a finite-horizon dynamic game, with each player represented as an autonomous agent whose policy depends on the current game state. MPE is used as a motivating solution concept rather than an exact equilibrium. To improve interpretability, we use Expandable Decision-Making States (EDMS) and decompose the Q-value into a successor-feature basis and a linear reward-weight vector. The value basis is estimated by linear TD initialization followed by nonlinear refinement. Using tracking and event data from 95 J1 League matches, we compare the proposed formulation with an independent reinforcement learning baseline. Because the two formulations define TD errors in different target spaces, TD MSE is used only for within-formulation consistency. With EDMS fixed, the independent baseline assigns the highest value to forward movement in 99.21% of evaluated off-ball states, whereas the most frequent direction under the proposed formulation accounts for 17.63%. Team-level average Q-values show a negative association with season-level expected goals for the baseline and a weakly positive association for the proposed formulation. Qualitative analyses illustrate context-dependent valuations of off-ball movements and defensive positioning. Overall, the proposed formulation produces more context-sensitive action rankings, although the comparison does not isolate the MPE-inspired component.

Recent neural power-flow solvers, including emerging foundation models, achieve accurate voltage predictions, yet such accuracy does not necessarily imply physically consistent solutions. Even small complex voltage errors can yield large AC power-balance residuals. We study this accuracy-consistency gap across PIGNN-GC, GridSFM, gridfm-graphkit, and LUMINA on realistic 2224-bus Great Britain network (GBnetwork) scenarios, with cross-grid evaluation of GridSFM over 31 systems. Using a singular value decomposition (SVD) basis fitted to training AC power-flow solutions, we find that neural prediction errors contain substantial components outside the dominant solution subspace. To address this mismatch, calibrated solution-subspace projection (CSP) suppresses off-subspace prediction components after train-only bias calibration, reducing Mean PB by 67.0%, 37.8%, 40.5%, and 68.9% for PIGNN-GC, GridSFM, gridfm-graphkit, and LUMINA, respectively, relative to calibrated predictions, while improving voltage-magnitude accuracy in all four models. These results identify output-error geometry as an important factor in physics-consistent neural AC power flow. Code: https://github.com/Kimchangheon/neural-acpf-error-geometry

Reward post-training of flow models either reweights the model's own samples under a KL penalty or a frozen reference, often for thousands of updates, or backpropagates the reward and moves every sample without checking that the move is worth its size. We introduce MEND, a reinforcement learning method built on proximal velocity matching. MEND caps rewards within each prompt group, so samples that already score well receive no move. Below the cap, it proposes moves along the reward gradient and accepts one only when its capped reward gain exceeds a quadratic displacement price. The model then regresses onto the resulting velocity targets, with no KL term, frozen reference model, or advantage weights. In 100 updates, MEND outperforms Flow-GRPO (about 4k updates) on five of six evaluators at the same distance to base-model images. Under an equal-budget protocol, it surpasses ReFL and DiffusionNFT at every evaluated update across four training rewards, reaching PickScore 24.03 versus 23.92 and 23.43, respectively. A 300-update three-reward run also surpasses the five-reward DiffusionNFT model on all three rewards it trains on. MEND is general and easy to adopt: it applies to any flow backbone with a differentiable reward.

LLM-driven evolution enables iterative model development, but two practical goals remain underexplored: finding model designs that transfer across related tasks and sustaining improvement when training is expensive. We introduce Population Evolution (PE), a collaborative, hierarchical framework that connects ongoing local searches through shared experimental evidence. PE evaluates code changes across related training instances and shares the results to guide subsequent proposals and promotion to larger training scales. For expensive targets, PE searches small training subsets and screens candidates through peer and intermediate evaluations before full-target training. We introduce RMD-Bench to evaluate both settings across ranking, watch-time prediction, RL algorithm discovery, and LLM/VLM pretraining. Compared with standalone evolution at matched source iterations, PE raises mean best local gains from 7.01% to 8.97% in ranking and from 2.84% to 3.85% in watch-time, while improving the best larger-scale outcome in all three joint-discovery families. In watch-time discovery, PE improves best larger-scale gains with four of five harnesses and all four proposers. On new recommendation datasets under shared target-side calibration, every evaluated PE design improves over the reference in mean performance. Under matched total GPU compute, completed LLM discovery runs yield a best relative accuracy gain of 2.48% and 13 successful candidates for PE, versus 0.92% and none for direct evolution. VLM loss reduction reaches 8.78% versus 5.05% under matched total GPU compute.

Site-specific fertilizer recommendation systems adapt nutrient advice to location, soil properties, crop type, and production targets, but scientific reuse is constrained when recommendation functions remain accessible mainly through interactive interfaces, outputs are not versioned, and trained approximations cannot be independently loaded or benchmarked. This technical report presents the Turba fertilizer machine learning stack, a three-layer open-source implementation for reproducible site-specific fertilizer recommendation in Morocco. \texttt{turba-client} provides programmatic access to publicly accessible site profiles, crop-specific target-yield spaces, and N, P$_2$O$_5$, and K$_2$O recommendation workflows; \texttt{turba-data} distributes analysis-ready snapshots; and \texttt{turba-models} packages crop-specific machine learning surrogates of recommendation outputs. The architecture links upstream retrieval, versioned analytical snapshots, reproducible cross-model benchmarking, and loadable offline surrogates while preserving the distinction between recommendation-system outputs, observed agricultural data, and model-generated predictions. The first dataset was constructed from 44,096 unique ESA WorldCereal locations. Scenario expansion across supported cereal workflows generated 132,017 crop-location recommendation requests under a medium target-yield setting. The resulting 22-variable dataset spans 10 regions, 66 provinces, and 1,149 communes. Nine regression families were evaluated under a fixed deterministic 80/20 protocol, and the current release packages five best-performing crop-specific models. The machine learning task is recommendation-function emulation rather than prediction of observed crop response. The stack provides a reproducible basis for spatial and temporal validation, uncertainty estimation, field-trial comparison, and future integration with additional data.

InnerSource (IS) is increasingly viewed as a way to improve large-scale software development. However, evidence on its adoption in public sector organizations remains limited. This paper reports a study of IS adoption in a Swedish Public Sector Organization. Drawing on nine semi-structured interviews and an internal workshop with 13 participants, we examine motivations, challenges, and improvement strategies for IS adoption. The results show that the main motivations are improving reuse, strengthening cross-team collaboration, and enabling faster and more cost-efficient development. Key challenges relate to a lack of an open culture and mindset, management of reusable assets (e.g., ownership, documentation, and code visibility), and ways of working (e.g., siloed practices and limited time for contributions). The study also identified improvements to address challenges, including fostering an enabling culture and mindset for IS and knowledge sharing, strengthening management support, and establishing internal platforms and governance for reusable assets.

Ontology documentation, visualization, and validation are usually carried out with separate tools. This split workflow slows down development and makes knowledge transfer harder. We present OntoInk, an open-source MkDocs plugin that brings these activities together. Within a single documentation-as-code pipeline, OntoInk renders interactive ontology diagrams, validates instance data against SHACL shapes, runs OWL\,DL reasoning, and supports inline Turtle editing. General-purpose diagram plugins for MkDocs cannot parse RDF, dereference IRIs, overlay SHACL constraints, or run OWL reasoning. Compared with standalone ontology visualization tools, OntoInk embeds interactive and editable diagrams directly into documentation pages. A live demo and source code are available at \url{https://ise-fizkarlsruhe.github.io/ontoink/}.

Agent skills package task-specific knowledge and procedures that can be composed to support complex agent tasks, while public marketplaces provide a growing pool of reusable skills. Existing security vetting, however, largely evaluates skills in isolation, leaving composition-induced risks underexplored. Such risks arise because composing benign skills expands the agent's capability space, enabling behaviors unavailable to any skill alone. Interestingly, we find that directly composing benign skills can already induce malicious behaviors, even when every individual skill passes security vetting. We further find that some target malicious behaviors remain difficult to realize through direct composition, even when the selected skills collectively provide the required capabilities. To systematically instantiate these attacks, we present Compositional Risk Induction via Multi skill Execution (CRIME). CRIME first uses the Malicious Plot Casting (MPC) module to decompose a target malicious behavior into complementary requirements and identify suitable benign skill compositions from public skill repositories. For compositions that cannot directly realize the target behavior, the Runaway Reaction Steering (RRS) module uses execution feedback to iteratively refine the selected skills toward the target while requiring each skill to remain benign under standalone vetting. The resulting composition is then passed to the Skill Reaction Chamber (SRC) module, where the skill pair is executed in a sandbox and the resulting environmental consequences are examined to determine whether the target behavior has occurred. Unsuccessful cases are returned to RRS for further refinement. Furthermore, we construct a benchmark of 4,000 public skills across eight cybersecurity behaviors for systematic evaluation of composition-induced vulnerabilities.

Business Process Management (BPM) was built on a foundational assumption that organizations are populated primarily by human actors whose work can be made visible, governable, and improvable through process models. That assumption is depreciating. AI agent ecosystems increasingly execute, coordinate, and adapt organizational work with limited human direction, challenging not only BPM's methods but its core conception of what a process is. We argue that BPM faces a constitutive shift from modeling human work to governing autonomous agents, for which we propose two new concepts: the \emph{Process Constitution}, a machine-interpretable, value-laden framework that defines the space of admissible agent behavior, and the \emph{Process Steward}, a governance agent that interprets and enforces it. The central value proposition of this new generation of BPM is not efficiency but \emph{organizational legibility}: the capacity to keep agentic organizations accountable, contestable, and humanly understandable. We outline what this means and sketch the research agenda it opens.

Despite their impressive performance on a wide range of video understanding tasks, current Vision Language Models (VLMs) are predominantly designed for offline scenarios and struggle to handle online streaming videos that demand low latency response. Several studies have explored memory and token compression strategies in an attempt to adapt offline VLMs for streaming video understanding tasks. However, through our probing experiment, we identify that most existing works tend to progressively lose long context information as length of input stream increases. To address this, we propose ReMem, a novel training-free adaptation technique that enables VLMs to process streaming videos of arbitrary lengths while improving their long context information retention capability. ReMem exploits memory from two perspectives, implemented as two core components. The Streaming Context Memory (SCM) continuously compresses historical context with query-independent attention. The Retrieved Vision Memory (RVM) then retrieves the most salient, query-relevant context from memory to augment the VLM's input. Comprehensive experiments demonstrate that the proposed ReMem achieves state-of-the-art (SOTA) performance across a variety of widely used benchmarks, spanning both streaming video and general long video understanding tasks.

One of the major transformative factors in 6G will be the integration of Artificial Intelligence (AI) to become a native part of Radio Access Network (RAN). While most physical-layer AI features have so far been evaluated in isolation using link-level simulations, their combined behavior in a realistic multi-cell, multi-UE deployment has remained largely unexplored. In this paper, we present system-level performance results when multiple uplink AI features are enabled together, achieved by integrating accurate link-level and system-level simulators. To infer state-of-the-art deep-learning-aided Multiple Input Multiple Output (MIMO) receivers under the dynamic allocations produced by a realistic uplink scheduler, we propose a mirrored data augmentation method that decouples receiver performance from scheduled allocation size. In addition to these Physical Layer (PHY) receiver features, we combine several recent advances in deep reinforcement learning to train uplink power control and link adaptation that outperform a heuristic baseline and further boost the gains obtainable from the AI receiver alone. The system-level results show that the combined AI features improve the mean uplink user throughput by roughly 27% compared to a non-AI baseline, confirming that the individual PHY and Medium Access Control (MAC) AI features provide complementary gains when deployed jointly.

Language models are moving beyond generating answers to pursuing long-horizon goals in interactive environments. Post-training these agents requires long, heterogeneous trajectories, and synchronous systems leave learner engines idle until rollout and verification finish. To squeeze out these pipeline bubbles, asynchronous training overlaps rollout and learning across updates, but comes at the cost of policy staleness. We introduce ThunderSyncRL, which starts gradient computation as soon as all required inputs are fixed, without policy staleness. For group relative policy optimization (GRPO), ThunderSyncRL computes each trajectory's score gradient as soon as the reward for that trajectory arrives, without waiting for the group. For on-policy distillation (OPD), it computes gradients for each completed agentic turn's teacher-scored actions while tool calls run in the sandbox. We prove that gradient streaming produces the same GRPO and OPD updates as batch-synchronous training, without changing either objective. On SWE-bench Verified and Terminal Bench 4.0, we train models to the same performance up to $1.9 \times$ faster than synchronous training. With zero policy staleness, ThunderSyncRL also outperforms asynchronous training at a fixed budget by up to $2.47$ percentage points.

Visual voice cloning requires intelligible, speaker-consistent speech synchronized with visible articulation. However, sequential multimodal conditioning can disrupt previously established temporal and speaker cues, while imbalanced inference guidance can improve linguistic accuracy at the expense of lip synchronization. In this paper, we propose UltraDub, a Unifying Visually-Steered Flow learning and trajectory Guidance Dubbing framework that leverages vision in two ways: as continuous motion for multimodal context aggregation, and as structural rhythm for trajectory rectification. Specifically, we introduce the Motion-guided Dual-context Retrieving (MDR) module, which continually recalibrates linguistic and speaker-style retrieval through shared lip-motion query residuals, utilizing independent time-conditioned gates to regulate their contributions. Furthermore, we propose Rhythm-anchored Trajectory Guidance (RTG), a training-free mechanism that evaluates hierarchical multimodal corrections at a visual-only predictive midpoint, safely strengthening semantic conditioning while better preserving temporal alignment. Finally, we construct DiverseDub, a multi-scenario benchmark to evaluate video dubbing in the wild. Extensive experiments demonstrate that UltraDub achieves state-of-the-art performance across four datasets.

Entropy-based selection objectives suffer from a fundamental degeneracy: minimizing Shannon entropy $H(p_A)$ rewards confident selection regardless of whether the selected candidate is informative. We address this limitation with the residual entropy $D = H(p_A) - H(p_β)$, where $p_β$ is induced by candidate trust weights. We prove the exact identity $D = -\mathrm{KL}(p_A\Vert p_β) - Δ$, where $Δ$ measures whether the score-induced distribution and trust profile favor the same candidates. Boundary cases establish basic safety: under uniform trust, $D\leq0$ automatically, so an equal-trust, non-starving state is never penalized, while at any one-hot limit, $D\to0$ regardless of the selected candidate. For the intermediate regime where selection occurs, we prove that $D\leq0$ when candidate ordering by trust agrees pairwise with ordering by informativeness, and derive a tighter certificate based on the leading candidate's margin over its competitors. These results are independent of the candidate-scoring function and apply to both stationary and dynamically changing information. Experiments with a gradient-based mixture-of-experts router confirm that the ordering conditions can hold during real optimization and show that correct ordering improves downstream performance when candidates are non-interchangeable and selections are used directly rather than averaged. Beyond routing, margin-based reweighting matches or outperforms fixed-strength baselines in a class-imbalance task, while informative selection in a production video-prediction system reduces MSE by approximately 20$\%$ and transfers to a related species. Residual entropy, therefore, provides a safety criterion for selection and a usable signal for deciding when that selection is informative.

Competent agents need precise and verifiable environments, such as sandboxes that are resumable at any stage and evolve from observable evidence. However, most long-horizon work exposes how rare these are: for example, an agent in medical research must ground a finding, classify it, and write a report over dozens of dependent steps, yet recent environments score only the outcome. To address the challenges in stable training, we present VERA, which builds such environments at scale and lets agents evolve on them. VERA builds these environments from initial trajectories: an agent writes rubrics, executable checks, a judge verifies each sandbox, and only those that pass enter the training bank. On these environments, VERA alternates between two updates: train the model with rubric rewards, or edit the harness skills. We also create a verifier which gates model checkpoints and harness edits using explicit development-set acceptance criteria. This attribution distinguishes VERA's co-evolution from single-axis baselines: its updates target not only the cause but the outcome. With an open-source corpus of 9,000+ long-horizon verifiable environments, a 9B model paired with its co-evolved agent beats the strongest baseline by 10.3 and 13.0 points in the two domains. At 27B, it surpasses the baseline on AutoCoWorkBench (71.6) and AutoMedBench (80.7), transfers to unseen workflows, and retains general capabilities.

Evaluating the causal effect of new features is a central goal for online platforms. While recent literature addresses limited testing traffic via centralized portfolio optimization, this perspective abstracts away a critical institutional reality: experimentation is operationally decentralized. The experimenters who develop new features also dictate which hypotheses to test, and they are typically rewarded based on empirical average treatment effects that are prone to upward bias. Left unchecked, this principal-agent conflict can severely erode platform value, a structural failure that conventional centralized levers, such as significance thresholds and traffic budgets, cannot resolve. By reframing experimentation as an incentive design problem, we demonstrate that two practical mechanisms, sample splitting and shrinkage, can effectively bridge this gap. Sample splitting aligns incentives perfectly at a bounded traffic cost, while shrinkage consumes no additional traffic and guarantees that interventions with negative expected effects are strictly unprofitable to field.

In generative models, Optimal Transport (OT) is used to improve Flow Matching (FM) by reducing noise-data coupling cost. However, different noise-to-output assignments can yield nearly equal costs, raising a key question. Is cost alone sufficient to guide coupling design? We address this question by separating transport cost from routing, i.e., the destination reached by each noise sample. We show numerically how FM and OT can differ in routing while remaining close in cost. We examine its consequences in learned neural FM. Using the exact FM routing as an oracle, we further construct a routing-aware training coupling and find that it yields a directionally consistent improvement in generation over a cost-matched, cost-only counterpart. Our findings highlight what cost minimization can overlook and motivate using both cost and routing to evaluate the design of OT-based FM couplings. Code will be released upon acceptance.

Seizure monitoring and detection are crucial for reducing the morbidity and mortality associated with seizures. Current epilepsy care, often involving expensive video-electroencephalography (VEEG) monitoring, requires specialized expertise and is limited to in-hospital settings, and intrusive in nature. Seizure diaries, on the other hand, suffer from unreliability due to under-reporting, leading to incorrect therapeutic decisions. Wearable non-invasive seizure detection may offer a more tolerable and feasible solution for long-term ambulatory monitoring. This study explores a wearable remote monitoring system utilizing a single wrist-worn accelerometer device and capable of detecting multiple types of seizures, including shorter duration events. We enrolled 79 patients under video-electroencephalography monitoring to wear accelerometer devices and collect data. Concurrent VEEG recordings were reviewed by board-certified epileptologists to produce annotations, including seizure onset, offset, and seizure type. Using this data, we constructed a deep neural network based on the time-series ResNet architecture, which could discriminate among seizure and non-seizure events. Our proposed approach achieved a seizure detection sensitivity of 95.65% and an overall false alarm rate of 0.15/24 hours during the evaluation, which spanned 5576 hours of total recording. Additionally, it resulted in an area under the receiver operating characteristic curve (AUC-ROC) of 0.98 and an area under the precision-re call curve (AUC-PRC) of 0.67 when averaged over 20 patients who experienced 46 convulsive seizures. These promising results suggest that the proposed seizure detection system can be effectively used for long-term ambulatory seizure monitoring. Future steps include validating our findings in larger datasets and assessing the utility of detection for additional seizure types.

This paper revisits the distributed learning problem for training a multinomial logistic regression model with the Federated Averaging ($\texttt{FedAvg}$) algorithm. We concentrate on a scenario with arbitrarily large stepsizes and heterogeneous update rules where the devices may perform a different number of local updates in each round. We show that, with linearly separable data, $\texttt{FedAvg}$ is stable with any stepsizes and the objective values converge to zero at the rate of ${\cal O}(1/R)$, where $R$ is the number of communication rounds. Our result also demonstrates that the effects of device heterogeneity vanish asymptotically. For sufficiently large $R$, the objective values decrease monotonically and is bounded by ${\cal O}( 1 / (R T_{\rm avg}))$, where $T_{\rm avg}$ is the average number of local update steps per communication round across devices. Numerical experiments support our findings.

Task-fMRI connectomes reveal state-dependent neural reconfigurations, yet conventional methods marginalize these signals by aggregating distinct conditions into static pairwise graphs, thereby obscuring condition-specific multi-ROI organization. We introduce CoHyFuse, a condition-aware ROI-centered hypergraph framework that constructs a task-state-specific incidence matrix from condition-wise functional connectivity (FC)-profile embeddings, allowing the same ROI to form different multi-ROI hyperedges across task phases. Condition-specific neighborhood sizes $K_q$ further adapt the hyperedge scale to each task state, and the resulting condition embeddings are fused with a complementary whole-session FC branch for prediction. In the AABC cohort (N=1,074), CoHyFuse achieved the best mean out-of-fold predictive performance among evaluated baselines on FACENAME Fluid Cognition Composite (FCC) prediction (7.83$\pm$0.10 MAE, 0.439$\pm$0.026 \(R^2\)) and VISMOTOR age prediction (7.52$\pm$0.37 MAE, 0.592$\pm$0.022 \(R^2\)). In an auxiliary CMI-HBN attention-deficit/hyperactivity disorder (ADHD) classification benchmark (N=223), CoHyFuse obtained 72.0$\pm$2.1\% macro-AUC and 74.2$\pm$2.9\% accuracy. Ablation studies support the contributions of condition-wise incidence construction and dual-view fusion, suggesting that state-resolved ROI-set structure provides complementary predictive information beyond whole-session FC alone. Occlusion analysis identifies the Distraction condition as the primary driver of model prediction, pointing toward the Salience/Ventral Attention Network (SAN)--FrontoParietal Network (FPN) and within-SAN hyperedge-defined ROI-set motifs as candidate model-relevant patterns. This framework provides an interpretable, state-resolved view of the connectome for downstream cohort analysis.

The last mile toward enterprise AGI is a company that runs itself. Training and adapting such agents require longitudinal enterprise data, which remain scarce, costly to acquire, and often restricted by privacy constraints. Historical archives are also frequently incomplete and record only what actually happened. They cannot show the outcomes of alternative decisions. We introduce MiniCorp, an office simulator for studying how agents can collectively run a company while generating enterprise data at scale. Using an e-commerce company as a demonstration, MiniCorp connects two interacting worlds. The external world models customers, dynamic competitors, and market mechanisms. The internal world consists of agents that observe events, discuss their options, and make strategic decisions. These decisions have lasting effects on the market, and the resulting feedback informs the firm's later decisions. As the firm and market interact, MiniCorp continuously records the agents' communications and decisions. These records preserve the information available at the time and the business results that followed. Checkpointing allows the same situation to be replayed under different decisions, providing comparisons unavailable in static archives. We evaluate end-to-end fidelity against patterns reported in empirical studies of real markets. These evaluations provide agents with realistic market feedback and reduce the risk that they learn to exploit flaws in the simulator. Our experiments show agents coordinating across roles and adapting their decisions to market feedback. With explicit long-term strategic guidance, they also sustain advertising exploration despite weak early returns. MiniCorp thus provides an environment for studying AI-run companies and a scalable source of longitudinal and counterfactual enterprise data for agent training and evaluation.

"Don't tune hyperparameters on the test set" is often stated in machine learning textbooks. Violating it is considered a cardinal sin that produces misleadingly optimistic results, corrupts benchmark integrity, and thus can even be interpreted as scientific fraud. Yet evidence suggests that test set hyperparameter tuning does occur in practice, making it all the more important to understand its actual consequences. So how bad is it, really? In this work we question this dogma and put it to an empirical test. We systematically study the magnitude of the performance inflation caused by tuning the hyperparameters on the test set for MNIST-1D, CIFAR-10, and three tasks from the GLUE benchmark. Our experiments show that while the effect is real and significant, it is frequently small relative to other sources of noise. In many cases, we find that tuning on the test set recovers exactly the same model as when tuning on the validation set. Most importantly, we find that the rankings of models remain essentially preserved after tuning on the test set and therefore that consistent test-set tuning may not invalidate benchmarks or model selection. Our results call for a more nuanced view of tuning hyperparameters on the test set, stimulating researchers to openly report test tuning.

Real-world users often exhibit highly heterogeneous preferences over multiple objectives for LLM responses. A lightweight aligner can tailor these responses to individual preferences, but scarce user-specific feedback makes personalized training difficult. Learning shared initializations across users can support few-shot adaptation. However, heterogeneous preferences and competing objectives cause gradient conflicts across users and within each user, hindering effective initialization learning. This raises a central question: \textbf{how can we collaboratively learn aligner initializations that support few-shot adaptation to diverse user preferences?} To answer this question, we propose \textbf{A}pproximate \textbf{P}areto \textbf{O}ptimality (APO). We first group users whose updates are compatible, so that their information can be combined with less interference. Within each group, we combine gradient descent with controlled ascent to coordinate competing objectives and move towards preference-specific points on the Pareto front. This produces an initialization that is close to the optima of the users in the group. We then iteratively refine it using updates from few-shot local adaptation, making it more effective for personalization. Furthermore, we establish conditional suboptimality bounds for a one-local-step collaborative update and characterize how initialization error affects subsequent stochastic adaptation. Experiments on Fed-ChatbotPA and UltraFeedback show consistent improvements over existing methods using only 20 local examples.

Training a small projector between a frozen vision encoder and language model is an established approach to multimodal learning. As the parameter count of language models scales dramatically, we revisit which vision capabilities this approach can add while keeping their pretrained weights fixed. Here we train a 50M parameter projector from the vision encoder of Kimi K2.6 to GLM 5.2 and 5.3, both models without native vision capabilities, and further present a reproducible recipe for training these adapters at scale. We study the following: (a) how vision capabilities of multimodal models scale as purely the language model side scales, and (b) what specific vision capabilities are able to be imbued into a pure language model at scale, and which ones remain limited. We evaluate on MMMU-Pro and BLINK, examining both overall performance and results on individual visual tasks.

Rapid progress in video generation has led to a plethora of models that differ substantially in capability and generation cost. This raises a natural question: can each request be efficiently routed to an appropriate model? We find that even when the consensus of the other annotators is used as an oracle, it agrees with each annotator's own favorite only 34-55% of the time. Motivated by this observation, we introduce TasteRoute, a personalized video-generation router that selects a generator jointly based on the input request, user preferences, and available generation budget. Across text-to-video and image-to-video settings, TasteRoute is competitive with strong simple baselines on preference routing while reducing average generation cost. The cost saving increases under higher budget caps. Finally, we release TasteRoute-3k, a human-annotated dataset containing multi-model video comparisons, quality judgments, preference rankings, and user-profile signals to facilitate future research on personalized and cost-aware video routing.

Masked diffusion language models (dLLMs) generate text by iteratively denoising masked positions, re-predicting each token multiple times before it is committed. An autoregressive decoder exposes an answer's distribution once, at the step that commits it; a dLLM exposes it at every denoising step before commitment, and we show that an adversary can exploit this. Since an answer remains open to revision over many denoising steps, an adversary with access to internal activations can watch how likely the model is to produce a chosen answer and adjust the intervention accordingly. Building on this observation, we study targeted bias injection, an attack that steers a frozen dLLM toward a demographic answer selected by the adversary. The attack uses a simple proportional-integral (PI) controller that tracks the target-answer probability during denoising and adapts the strength of a steering vector on the fly. On ambiguous BBQ questions where the correct answer is abstention, our attack raises LLaDA-8B-Instruct's preference for the targeted group from 1.8 to 16.7 percentage points, more than three times the strongest fixed-strength steering baseline, and on SocialStigmaQA it raises the selection of stigmatizing answers from 17.6% to 58.1%. Fitted to other demographic targets, the same attack shifts answers by up to 37 percentage points, and each attack takes about 40 minutes on one GPU. On the primary target, feedback is what makes the attack work: constant steering at the same average strength over the token-committing steps produces a far smaller shift while corrupting nearly three times as many outputs, and a constant strength set separately for each example still falls well short. Our findings identify the denoising trajectory as a new control channel in dLLMs and call for bias audits that examine the serving stack rather than the frozen model alone.

Many works approach continual learning through the lens of infinite context windows. As an agent puts more observation into context (concretely the KV cache), compacting said context is akin to direct memory manipulation, without affecting the base model's weights. Many works pose KV compaction as an optimization problem: learn a smaller set of KV vectors that matches the behavior of the full KV cache. While this preserves base model behavior, optimizing through a frozen base model results in a highly nontrivial optimization problem with a brittle and flat loss landscape. In this paper, we characterize what makes these optimization problems difficult and demonstrate that a heavily simplified Perceiver-based architecture not only matches performance of a full Perceiver transformer in continuous context compaction, but outperforms baselines on compaction utility. Results are presented on MCQ tasks across Finance, Legal, Gutenberg, and Code.

Learning from human demonstrations is a reliable way to teach robots new tasks, but the gains from each additional demonstration shrink as the policy improves. Continued improvement can instead come from supervised deployment, where an operator places the objects and intervenes when the policy fails. We ask how to maximize improvement from a fixed budget of supervised episodes on high-precision manipulation tasks with wide ranges of object placements. We observe that failures can concentrate in a small subset of initial states, so uniform collection spends much of the operator's time on states the policy already handles. Mulligan makes the initial-state distribution a decision, starting each round's episodes at observed failures and untried states. To further improve data efficiency, we augment interactive imitation learning with a value function trained on all data, including failures that imitation discards. Across three real-world tasks evaluated on 2,550 held-out, blinded episodes and two simulated tasks, Mulligan outperforms uniform initial-state sampling at matched collection budgets, and combined with value-based action selection, HiL-IDQL+Mulligan, improves final real-task success by 10-34 percentage points. With operator interventions, the human-robot team completes 98% of collection episodes, remaining productive while the policy learns. Videos, code, and data are available at https://mulligan.page/.

We present the Prior-Fitted Language Model (PFLM), a 300M-parameter byte-level transformer pretrained only on samples from a synthetic non-linguistic prior. Given a prefix of real text, it learns to predict the language in context with frozen weights, having never seen a word of any real language. Every training sequence is generated by a recurrent structural causal model drawn fresh from a distribution over such models. The model never sees the same language twice during training, so the only way to predict the continuation is to infer the language from the prefix. Samples from this prior share the statistical signatures of natural text: Zipfian frequencies, slow entropy-rate convergence, and long-range dependence. On Wikipedia in six languages, bits per byte fall from the uniform eight to between 0.9 and 2.4 at one million bytes of context. Given numerals instead of text, PFLM learns to count, to compare magnitudes, and to add approximately. It predicts deterministic sequences like Rudin-Shapiro or the prime indicator, and it compresses six non-text domains, from source code to speech, below gzip and PPMd. The model has not learned a language. It has learned to learn one.

Benchmarks expose vision-language-action (VLA) policies to few canonical instructions, while exhaustive deployment testing is impossible. We introduce Object-Grounded Attention Monitoring (OGAM), connecting systematic testing to runtime assurance: testing reveals attention divergence between successful and failed executions, and OGAM uses this signal to stop failures beyond the finite suite. We generate scene-grounded instructions through pairwise combinations of action templates and objects, and separately test meaning-preserving paraphrases. All 87 out-of-benchmark cases reveal problematic behavior across OpenVLA, OpenVLA-OFT, UniVLA, and $π_{0.5}$: none completes any of the 24 feasible instructions, while infeasible or hazardous requests also trigger behavior substitution. At each action query, we project gradient-weighted visual attention through object masks and group it by instruction role for comparison across tasks and policies. Dynamic time warping aligns this course with a successful reference despite speed differences; conformal calibration on successful episodes sets the early-stopping threshold for sustained deviations, with a nominal false-stop target of $α=0.05$. Across four policies, OGAM stops 87-100% of failed episodes at median times of 5-12s within a 20s budget, with observed false-stop rates of 3-5%, without failure-labeled training. Finite testing thus identifies attention patterns that support online intervention before failure fully unfolds.

Scalable Graph Transformers are commonly trained and evaluated on static large graphs in a transductive setup. Many scalable Graph Transformer components can be formulated as a constant-size shared memory, similar to virtual nodes, providing compressed information about the whole graph. The counterpart of these models in language models and other domains is justified as the input changes, and this mechanism learns to compress some useful information about the input. In transductive learning on a single fixed graph, however, any shared memory can be seen as a constant at test time. This raises the question of what exactly this shared memory does in this static setup. We give preliminary evidence that optimizing a shared memory directly performs similarly to global communication methods, and so normal local message-passing models can embed similar information in their weights. Thus, these settings may be a poor fit for evaluating global communication in graph neural networks.

A long-standing goal of AI is a model that can continually learn and improve itself. On post-trained models, supervised finetuning (SFT) on new data often causes poor generalization and catastrophic forgetting. As such, the conventional wisdom is that on-policy training is a prerequisite for continual learning. In practice, however, data containing new knowledge or capabilities are often off-policy. While methods such as on-policy self-distillation (OPSD) try to bridge this gap by converting off-policy data into on-policy signal, they have been shown to cause reasoning collapse. In this paper, we show that off-policy merging beats OPSD for continual learning. We first show that SFT learns a useful signal from new data, but naively applying its update interferes with existing capabilities. We reduce this interference with a simple recipe we term grafting, which changes where the update is learned and how it is applied: (1) learning the update on an earlier donor checkpoint, ideally even before the end of pretraining, and applying the weight update to the post-trained model; (2) scaling the weight update, equivalent to a form of model merging; and (3) optionally, masking the most sensitive update directions when the new data distribution is far from the post-trained model. Across continual learning settings including (1) distilling from expert traces, (2) self-improvement with STaR and Pedagogical RL, and (3) injecting knowledge after pretraining cutoff, grafting Pareto-dominates both SFT and OPSD in new-task and old-task performance, while avoiding expensive on-policy sampling. Therefore, our work challenges on-policy training as a necessity for continual learning on RL-trained models.

This paper studies online quantile regression for large-scale and streaming data using Stochastic SubGradient Descent (SSGD) with constant learning rates. Classical offline inference for quantile regression is computationally and memory intensive. Existing works of online inference for quantile regression provide only asymptotic guarantees and typically require sub-exponential tail conditions for distribution theory. To bridge these gaps, we introduce new techniques to prove a quenched central limit theorem (CLT) and finite-sample Gaussian approximation for SSGD under a finite-moment assumption. We further show that Ruppert-Polyak averaging with a constant learning rate has a non-vanishing bias and fails to satisfy CLT centering at the population target. Hence we propose suffix averaging to address this issue and establish its finite-sample Gaussian approximation. Based on these results, we provide an efficient online inference method for quantile regression that avoids covariance estimation. Numerical experiments show that our method achieves desirable empirical coverage rates and competitive performance compared to other inference methods. We also apply our approach to U.S. wage data to demonstrate its practical effectiveness.

Feature pyramid methods, from FPN to BiFPN, have achieved strong performance in face detection by fusing multi-scale features. However, detecting faces under unconstrained conditions, such as small scale, occlusion, and extreme pose, remains difficult, as it requires global cross-scale dependencies that local fusion cannot model. State space models such as Mamba provide global context with linear complexity by scanning features as a sequence, and therefore offer a promising direction for this problem. Nevertheless, such a scan needs the two pyramid scales combined into a single feature map, and the way they are combined determines whether cross-scale structure is preserved. Summation collapses the two scales before the scan, so the scan has no cross-scale structure to exploit, while concatenation keeps both scales but at far higher cost. To address this, we propose \textbf{Weave Mamba Fusion (WMF)}, which interleaves two adjacent pyramid scales column by column so that each step of a horizontal bidirectional SS2D scan moves from one scale to the other. With partial-channel processing and parameter-free de-weaving, WMF enables efficient cross-scale interaction while preserving feature structure. Integrating WMF into every fusion node yields \textbf{WeaveBiFPN}, the neck of our \textbf{WeaveFace} detector. On WIDER FACE, WeaveFace achieves 91.41\% mean AP with only 0.34M parameters and 1.16 GFLOPs, outperforming prior detectors under 0.5M parameters. Its largest gains are on the Hard subset, where it reaches 87.14\% AP. The code is publicly available at \url{https://github.com/dohun-mat/WeaveMambaFusion}.

High-quality child care in early life is a critical determinant of children's growth and development. Research on child care quality has been constrained by fragmented, non-research-friendly, and privacy-bound datasets. We present CCQ (Child Care Quality), a large-scale, de-identified dataset for applied data science research at the intersection of AI and early childhood health. CCQ integrates 59,372 child care provider records across 12 U.S. states, covering diverse provider types as well as data schemas. To ensure research utility while protecting privacy, we implement an automated, LLM-based curation pipeline that anonymizes, cleans, and standardizes raw state records into two complementary releases: a cleaned textual release and a fully preprocessed tabular release. We also benchmark traditional machine learning models, tabular foundation models, and language models on quality rating prediction and important features analytics. Within a state, tabular classifiers on the preprocessed tables perform best. Across states, zero-shot transfer is near chance, but modest target-state supervision recovers most of the within-state performance, and pretraining on other states benefits finetuned language models. We release both datasets with all code to accelerate AI-driven research on child care quality and ultimately improve children's health and development.

Graphical User Interface (GUI) agents have emerged as a promising paradigm for automating complex digital workflows across diverse applications. However, training highly capable and generalizable agents fundamentally relies on massive, high-fidelity visual-action trajectories, which are notoriously difficult to acquire. While human demonstrations are unscalable, existing GUI world models rely on text descriptions or HTML rendering, discarding crucial pixel-level visual details like icons and layout styles. To address this issue, we introduce Infinite-Dreamer, a simulation-free data synthesis method powered by a pixel-level Image Editing World Model. By conceptualizing GUI transitions as image editing tasks, we leverage Vision-Language Models (VLMs) to describe action-induced UI changes as structured delta-text. We then fine-tune an image editing backbone to controllably synthesize realistic screenshot transitions. We utilize this model to generate both single-frame visual robustness data and multi-step imaginary trajectories. To validate the effectiveness of our approach, we fine-tune the Qwen3-VL baseline solely on the synthesized data to obtain Infinite-Actor, and evaluate it on AndroidWorld, MobileWorld, and AndroidControl-Curated benchmarks. Infinite-Actor consistently outperforms the Qwen3-VL baselines across scales: Infinite-Actor-8B improves AndroidWorld Pass@1 by +4.45 and nearly doubles the MobileWorld Pass@3 success rate, while Infinite-Actor-2B improves Pass@1 by +9.05. Code is available at https://github.com/swaydy-n/Infinite-Dreamer.

Cognitive Diagnostic Models (CDMs) identify which specific skills a student has and has not mastered, the signal a personalized learning path needs and a single aggregate score cannot give. Yet they are rarely deployed. The obstacle is their precondition: the Q-matrix, a mapping from every assessment item to the skills it requires, historically authored by hand. We separate the task into two stages: first construct the curriculum's own knowledge graph, the full space of concepts and skills it contains, independent of any item; then map items against that graph on demand. This paper addresses the first stage only. The item-mapping stage is designed but not implemented here, so the claim that this shifts judgment cost from once per item to once per curriculum is a design rationale rather than a finding. We present Curriculum Brain, a two-repository system pairing a version-controlled knowledge base with an agentic pipeline of eleven single-responsibility agents under a thin deterministic orchestrator. It generates candidate concept-skill mappings from official curriculum documents, checks them against accumulated rules, and compares them with a concept-skill map extracted independently from the textbook, repairing its own failures and escalating to a human only when it cannot resolve a case itself. Across 241 chapter runs (168 distinct chapters), 41.5% produced a Generator output passing both checks without a patch, and 67.6% resolved without escalation. Both are measured against criteria the system itself produced, so both describe internal consistency rather than agreement with an external standard, and both pool two pipeline configurations separated by a single change at run 77; after it the figures are 57.0% and 91.5%. Observed spend was $1.19 per chapter, API spend only, excluding human review. We release both the framework and the resulting curriculum dataset.

As large language models (LLMs) become increasingly capable of code generation, adopting generated code in software development requires assessing not only its functional correctness but also its maintainability-related quality. If such quality could be estimated before generation, developers could avoid the cost of generating, reviewing, and discarding low-quality code. Although prior work has shown that the functional correctness of the LLM-generated code can be predicted in advance, it remains unclear whether maintainability-related quality is similarly predictable. We introduce Pre-Generation Maintainability-Related Quality Prediction (PreMaQ), which predicts the Code Smell Score (CSS) and Maintainability Index (MI) of generated code from the internal representations of LLMs before generation. Our evaluation covers four open-weight LLMs and four Python code generation benchmarks, comprising 2,695 tasks in total. Our results show that predicted CSS and MI consistently correlate with their observed values across all 16 model-benchmark combinations, achieving mean Spearman rank correlations of 0.57 and 0.65, respectively. When used for model selection, PreMaQ achieves 59.5% of the maintainability-related quality improvement attainable by an ideal maintainability-based selector over random selection on tasks for which multiple models generate functionally correct code. Combining predictions from PreMaQ and prompt embeddings increases this proportion to 60.7%, indicating that the two signals are complementary. These findings suggest that PreMaQ can be used to predict and improve the maintainability-related quality of LLM-generated code.

A bipedal robot cannot deviate from its path to avoid an obstacle without disturbing its balance, and this coupling is most severe on underactuated platforms such as the biped considered here, which has four actuated joints per leg and no hip or ankle roll. This paper presents a Hierarchical Reinforcement Learning (HRL) framework in which a High-Level (HL) policy observes the robot pose, 36 raycast proximity measurements, moving-obstacle states, and a receding-horizon local goal, and outputs a body-velocity command $(v_x, v_y, ω_{yaw})$ every ten control steps, while a velocity-conditioned Low-Level (LL) policy tracks each command through PD-controlled joint targets. Both policies are trained jointly with Soft Actor-Critic (SAC). Because the converged gait is task-agnostic, it is frozen and driven by classical planners over the same command interface, yielding three controlled baselines: SAC+A*, SAC+RRT*, and SAC+APF. Across 100 evaluation trials per method in randomized PyBullet environments, the proposed method reaches the goal in 98.0% of static and 88.0% of dynamic trials, against at most 78.0% and 68.0% for the planner hybrids, with path lengths within 4% of the A* reference, and ablations confirm that each observation channel and reward term contributes materially to this performance.

Triangular attention is a core computation in AlphaFold3-style biomolecular models, with cubic cost in token count. Its shared pair bias adds a gradient reduction across attention slices to the usual reductions over queries and keys. The open-source backends we examine handle these reductions through repeated probability recomputation, floating-point atomics, or full score-gradient storage. Separately, computing the softmax backward correction from BF16-rounded forward outputs loses numerical precision. We present TurboPairFormer, a triangular attention implementation for NVIDIA Hopper GPUs that addresses both issues. Our key-tile-parallel backward algorithm recomputes each probability tile once for the query, key, value, and pair-bias gradients, using ordered partial reductions for deterministic accumulation without floating-point atomics or full score-gradient storage. Output-residual compensation retains a BF16 approximation of the output-rounding residual to compute the backward correction more accurately in FP32, without changing the BF16 output. With BF16 inputs at crop sizes 384, 640, and 768 and head dimensions 16 and 32, TurboPairFormer achieves the lowest mean query, key, and pair-bias gradient RMSE against an FP64 reference among the implementations compared in this paper. Residual compensation reduces these RMSE values by 28-47% in controlled ablations. All four gradients are bitwise identical across five repeated calls in all 600 input cases under fixed execution conditions. Integrated into OpenFold3 with our triangle multiplication kernels, TurboPairFormer achieves the lowest GPU computation time per optimizer step among the evaluated backend configurations on 16 H100 GPUs, with speedups of $1.73\times$ over OpenFold3's Triton backend and $1.13\times$ over cuEquivariance at crop size 768.

Monitoring concept drift from an adaptive classifier's error stream creates an operational conflict with the model's own update loop. When internal adaptation outpaces evidence accumulation, accuracy recovers before cumulative detectors (CUSUM, Page-Hinkley) can reach threshold. Instrumenting an Adaptive Random Forest (ARF) shows that surviving trees absorb 98.6% of the post-drift error transient through incremental leaf updates alone. The first background tree swap accounts for just 0.71% of this erased error volume, but drops external detection rates by 31 percentage points. We derive the finite-horizon boundary where cumulative evidence fails to cross threshold and measure a critical magnitude floor ($Δe_c = 0.120$) below which false-alarm budgets preclude detection. This failure manifests as missed shifts on stationary streams and false-alarm flooding triggered by internal tree swaps on noisy baselines. We validate on synthetic shifts, ARMA-GARCH series (ProteuS), and tabular benchmarks (BAF, INSECTS); on the synthetic sweep at a standard threshold, the blind spot appears at $Δe \approx 0.25$, showing why classical benchmarks like SEA ($Δe \le 0.21$) failed to reach it.

Frozen representations are widely reused for downstream classification, yet each new task typically requires fitting a new predictor. We ask whether the few-shot prediction procedure itself can instead be learned once and reused across datasets and representation spaces. To study this question, we introduce RepShiftBench, comprising 1,218 encoder--dataset tasks across text, image, and audio, with separate evaluation of generalization to unseen datasets, unseen encoders, jointly unseen datasets and encoders, and unseen modalities. The benchmark exposes a substantial gap: Logistic Regression fitted independently on each episode outperforms every evaluated in-context learner across all settings. We introduce RepICL, a meta-trained in-context learner that canonicalizes each episode through episodic whitening before prediction. Its inductive variant, RepICL-I, surpasses Logistic Regression in all 12 benchmark settings, while RepICL-T substantially outperforms existing transductive methods. Ablations identify episodic whitening as the primary source of these gains, while showing that it is not a universally beneficial preprocessing step. Across both variants, the gains concentrate on queries for which simple support prototypes favor the wrong class or provide little separation between the true class and competing classes. Transduction provides its largest additional gains when limited support coverage gives a misleading view of class separation. Together, these results demonstrate that a shared few-shot prediction procedure can generalize beyond the representation spaces observed during training.

Diffusion models generate samples by learning to reverse a fixed corruption process, and classifier-free guidance (CFG) is the standard mechanism for conditioning this process on a desired class or prompt. CFG can be applied at varying guidance strengths, and while higher strengths improve image quality and conditional alignment, too high a guidance strength can degrade image quality and diversity. Furthermore, CFG violates principled diffusion sampling dynamics, and existing explanations for why it works despite the violation disagree on the underlying theory or do not extend to deterministic samplers used in practice. We address both these issues. We first frame CFG sampling as a continuous-time optimal control problem, treating the sampling trajectory as a sequence of controls chosen to maximise the probability of the desired condition. Solving the resulting Hamilton--Jacobi--Bellman equation shows that CFG is recovered under specific path costs when using an unconstrained control set. We argue this lack of constraint is responsible for CFG's failure at high guidance strengths, since it permits the sampling path to move arbitrarily far from the current image estimate. To fix this, we propose curvature-aware CFG (CACFG), which constrains the control set to a hypersphere informed by the Gaussian regularisation used when training variational autoencoders. We show that the control inputs produced by CFG sampling routinely violate this bound, and that across diffusion models, datasets, and guidance schedules, CACFG achieves superior generative quality at mid-to-high guidance strengths with a less severe quality-diversity tradeoff than regular CFG.

Recovering complete phase sets from powder X-ray diffraction (PXRD) is challenging when weak-phase peaks overlap stronger signals. A natural strategy is to identify phases iteratively, removing the contribution of each identified phase from the observed pattern before predicting the next. However, even after a phase is correctly identified, misestimating its contribution can distort the residual and cause subsequent errors. We introduce PhaseMatcher, an autoregressive framework for complete phase-set identification with physics-guided spectral decomposition. After each phase prediction, PhaseMatcher re-estimates the contributions of all selected phases and the residual from the original observation and all selected reference patterns, accounting for physically plausible variation between reference patterns and the corresponding phase contributions in the observation. The resulting residual guides subsequent phase identification, while a separate stopping module determines when the phase set is complete. On synthetic mixtures and controlled mixtures constructed from measured single-phase patterns, PhaseMatcher improves complete-set identification over the evaluated baselines. On PhaseMix-135K, it also estimates contributions and residuals more accurately than scalar subtraction.

Predicting metal-organic framework (MOF) structures from given building blocks requires recovering their positions and orientations in a periodic crystal. The spatial effects of rotation errors are geometry-dependent and anisotropic. The same angular error can produce different atomic displacements depending on block size, shape, and rotation axis. Angular error alone, without reference to the specific block geometry, therefore cannot fully describe the spatial consequences of a pose error. We introduce AnchorPose, a meso-grained pose generation framework that incorporates this geometric dependence into its generative representation. It represents each block through a small set of representative atoms, combines their local geometry with the current spatial state, and generates their coordinates with Bayesian Flow Networks. Known atom correspondences enable rigid alignment to recover complete building-block poses and return geometrically consistent points to the generation process. This design connects point-level spatial prediction with block-level structural constraints. Geometry participates in the pose state and its prediction, while rigid reconstruction preserves intra-block structure without treating all atomic coordinates as assembly variables. On the MOF benchmark, AnchorPose improves single-candidate match rates over the compared block-level and all-atom baselines.

Linear attention enables efficient long-context autoregressive decoding by compressing history into recurrent states, but this compression can make selective access to sparse and distant information difficult. Existing chunk-based extensions increase memory capacity, yet learned chunk-mixing coefficients may remain fixed with respect to input content and therefore cannot adapt historical access to each query. We introduce \emph{Hybrid Linear Attention} (HLA), a query-dependent chunk-level attention mechanism for Gated DeltaNet (GDN). HLA represents each completed chunk as an exact affine state transition and computes content-dependent routing gates from compact, self-attentively pooled representatives. Each gate interpolates the corresponding historical transition with the identity map, controlling both the chunk's additive memory and its transformation of earlier states. Effective-support regularization further encourages concentrated routing for sparse inference. We evaluate HLA under both pretrained adaptation and from-scratch training. Across Qwen3.5 models from 0.8B to 9B, HLA consistently improves over native GDN and fixed chunk mixing, with gains of up to 5.57 percentage points on LongBench-V2 and 3.97 points on RULER. In a controlled from-scratch 1.3B setting trained for 100B tokens with a 4K context, HLA also improves RULER performance from 4K to 32K, with gains increasing from 0.83 points at 4K to 4.22 points at 32K. These results demonstrate that query-dependent composition of recurrent memory improves long-context modeling and remains effective beyond the training context while using compact per-chunk affine summaries. Project page: https://caesarhhh.github.io/hla/

Tactile Internet (TI) security analytics must balance reliable thresholded decisions with constrained computational and measurement resources. We study this tension for finite-shot hybrid quantum anomaly inference and introduce the Adaptive-Shot Variational Quantum Circuit (AS-VQC) policy. This validation-calibrated policy begins each record at 128 shots and cumulatively escalates through 256, 512, and 1024 shots only when the finite-shot anomaly score remains close to a validation-selected security threshold. The quantum scorer is evaluated as an off-path security analytics component rather than part of the haptic critical path. Using a 4,875-record CESNET-TimeSeries24-derived aggregate-flow benchmark, leakage-safe random, entity-group-disjoint, and temporal holdouts, and five trained quantum neural network (QNN) checkpoints per holdout, the primary AS-VQC-95 (beta = 0.95) policy averages 129.2, 276.9, and 131.2 shots per record, saving 87.4%, 73.0%, and 87.2% of the uniform 1024-shot baseline (Fixed-1024), respectively. The decision disagreement with analytic (exact-expectation) inference is 0.771%, 0.409%, and 0.635%, lower than both the uniform 128-shot baseline (Fixed-128) and a matched-budget shuffled-allocation control. Fixed-1024 remains more decision-stable, establishing a measurable reliability-resource trade-off rather than cost-free equivalence. A more conservative AS-VQC-99 (beta = 0.99) further reduces disagreement while using fewer than 512 average shots across all holdouts. These results show that finite quantum measurements can be treated as an inference resource and concentrated on boundary-sensitive TI-security decisions while exposing checkpoint-dependent escalation under unseen-entity conditions.

Broad Learning System (BLS) supports analytical training and incremental expansion, but its growth needs guidance on which inputs new blocks should learn from. Weight errors pose a further challenge by displacing learned outputs across class boundaries. We propose HiER-BLS to couple hierarchy-guided representation growth with error-correcting learning. Successive blocks focus on inputs selected by feature importance and correlation while preserving earlier representations. The evolving branch guides encoded learners through subspace size and sample confidence, so its learning experience informs both their feature views and supervision. For finite broad readouts, we show how codeword correlations transform fitted class scores. Prediction preservation depends on the distance from the actual output to the nearest decoding boundary relative to the model's sensitivity to weight errors. Experiments on five image and five tabular datasets demonstrate improved classification performance over representative BLS variants. Component studies show that hierarchy guidance benefits the encoded branch even when the guiding branch has lower standalone accuracy, with further gains from combining their scores. Longer codes continue to improve accuracy under stronger Gaussian weight errors after clean accuracy has largely saturated.

Long-running agents repeatedly call an LLM while retaining most of their document window, evicting old documents, and appending new ones. These rolling updates break exact prefix caching and motivate non-prefix KV-cache reuse with selective recomputation. We show that persistent KV-cache reuse with selective recomputation can be history-dependent: in our rolling-agent workload, an unchanged prompt can produce different answers depending on the requests processed before it. At a matched 5% recomputation budget, document-aligned recomputation reduces answer variation across request orders from 69.0% with CacheBlend's token top-$k$ policy to 26.1%. When each prompt is evaluated after a different sequence of preceding requests, document-aligned recomputation improves fidelity to full prefill by 34.5-52.5 percentage points over token top-$k$, while both policies achieve approximately 5.7$\times$ median TTFT speedup. Our ablation study shows that, in our rolling-agent workload, contiguity is the main factor associated with robust selective recomputation.

Terminal agents are commonly trained by imitating long teacher trajectories, yet how much of each trajectory to supervise remains unexplored. We study the \emph{supervision horizon}, the number of trajectory tokens retained for training, and show that it is a key design axis for reliability and cost. Reliability improves with longer horizons but saturates: on Terminal-Bench, a 12K-token horizon solves more tasks than 16K ($29\pm0.7$ vs.\ $26\pm0.8$) while requiring 30\% less training time. The horizon also shapes agent behavior: short horizons cause premature termination, intermediate horizons yield productive error recovery, and long horizons induce over-persistence. We analyze this saturation through a bias--complexity bound, in which longer supervision reduces temporal supervision bias but increases finite-sample estimation error from more heterogeneous late-stage histories. Guided by this analysis, we propose \emph{selective long-horizon refinement}, which first trains on short prefixes and then refines only on continuations that are most likely under the warm-start model. It consistently outperforms full long-horizon training. At 16K, it raises successful attempts from $110\pm2.7$ to $126\pm2.1$ and tasks solved in at least six of eight attempts from $9\pm0.7$ to $14\pm0.6$; with half of the long-horizon data, it still reaches $122\pm2.4$ while cutting training time by 23\%. The gains transfer across benchmarks, from $64\pm2.6$ to $73\pm2.1$ on Terminal-Bench v2.0 and from $137\pm2.7$ to $155\pm2.2$ on OpenThoughts-TBLite. For long-horizon supervision, selecting the right trajectories matters more than training on all of them.

Agentic leaderboards increasingly evaluate systems on public benchmarks whose task statements and solution-bearing artifacts can remain accessible. We propose a measurement-first audit framework that distinguishes contamination claims according to the evidence required to support them. It separates three channels that require different evidence: training-time exposure, evaluation-time retrieval, and pipeline/scaffold leakage. Each channel is coded as open, partial, closed, or unknown under a fail-closed rule. Across nine Holistic Agent Leaderboard (HAL) configurations, none of the 27 channel assessments was coded closed, but incidents were confirmed in four configurations. We then apply the behavioral component of the framework to a reported file-localization gap on SWE-bench Verified, using an outcome-blind, same-repository matched-control design with symmetric prompt-leakage screening, paired and repository-aware uncertainty analyses, and scorer validation, evaluated on GPT-4.1 and DeepSeek-V4-Flash. Among the 100 pairs retained after symmetric screening and the pair-integrity exclusion, GPT-4.1 showed a $+10.0$-point pair-weighted Top-3 benchmark-associated gap, but the 95\% intervals from both the prespecified paired-bootstrap procedure and the post-hoc repository-balanced analysis included zero, leaving the benchmark-associated gap inconclusive. The reproduction scorer did not pass its validation gate: against consensus human labels, sufficient scorer sensitivity could not be established for either model, and both DeepSeek-V4-Flash firings on correct-gold comparisons were false positives. Without provenance evidence, appropriate controls, symmetric leakage screening, and validated scorers, stronger contamination claims are not warranted. The results do not establish training-data membership, contamination prevalence, or benchmark-induced score inflation.

However, many existing steering approaches rely on target-domain human data for fine-tuning or prompting that is costly to collect and raises privacy concerns. In this paper, we study demographic group-level survey simulation, where personas induced from heterogeneous, anonymized public behavioral data condition agents that simulate responses of individuals from specific demographic groups. We examine whether representative personas can be induced from diverse sources and analyze how the domain, scale, and granularity of the source data affect survey simulation alignment. We find that personas induced from out-of-domain sources rarely outperform simulations conditioned only on basic demographic information, largely due to population mismatch. However, when personas are accurately assigned to the target demographic groups, alignment improves substantially. Finally, personas induced from target-domain survey data generalize better as more survey question history becomes available, suggesting that richer behavioral evidence enables more stable persona trait inference that transfers to better unseen questions simulation alignment.

Classification problem in the context of highly imbalanced data is a major challenge in many real-world applications (e.g., FinTech, healthcare, etc.). In these cases, the vast majority of instances belong to a single class and a small fraction represent the minority class (often the most critical class). Recently, diffusion models have emerged as powerful approaches to reduce the degree of ``imbalanced-ness'' in the dataset; they work by generating synthetic data by capturing complex data distributions using iterative transformations. However, standard diffusion models are not inherently suited to highly skewed or heavy-tailed data, due to inbuilt quadratic error loss, which lacks the structural sensitivity to capture rare, extreme values, and minority-class nuances. We propose a novel approach, namely, Quantile-TabDDPM, based on a quantile-regularized denoising objective that combines the standard quadratic error loss with a quantile loss term to explicitly capture rare events while preserving the theoretical grounding of the original denoising objective. We extensively evaluated our approach on a real-world credit card transaction dataset characterized by extreme class imbalance. The results demonstrate that the integration of diffusion-based synthetic data generation with a quantile-regularized denoising objective provides a robust and effective framework for fraud detection in highly imbalanced datasets.

Personal AI assistants are beginning to transact for consumers, and early adopters capture real savings. Whether those savings survive, and what happens to consumers who have no assistant, depends on how sellers respond -- a question single-user evidence cannot answer. We build an agent-based rental market in which language models play consumers, assistants, and six adaptive sellers guided by an algorithmic pricing tool. Half the population receives an assistant under an advisory or an executing mandate; the contract pairs a fee only the renter's physical action avoids with a pre-selected add-on an authorised assistant can cancel online. An analytical benchmark and a behaviourally calibrated rule market supply ex-ante predictions, and paired branches with frozen versus adaptive sellers separate adoption effects from market feedback. Across thirty simulated markets, executing assistants cut adopters' spending by 13.7 USD per renter-day when sellers are frozen; adaptation claws back about a third, leaving 8.7, with the gains arriving both as lower bills and as rentals completed at all. Sellers raise headline rates while cutting fees, and the calibrated forecast of the burden on unassisted consumers (+3.6) does not transfer: their mean spending change is +0.4, confidence interval -0.6 to +1.3. Seller-model swaps and a within-market transfer of fee-setting to the pricing tool show that fee conduct, and with it the division of the gains, is decided on the seller side. Assistants, we conclude, should be evaluated at market level -- completion, total spending, and non-users included -- and the comparison layers locate exactly where a calibrated behavioural forecast fails in a language-model market.

Transporting unsecured payloads with legged robots over uneven terrain requires balancing locomotion performance and payload stability, since aggressive motion can destabilize the payload even when the robot remains stable. We study quadrupedal transportation of unsecured stacked boxes on an edgeless torso-mounted board without dedicated payload sensors or active carrier mechanisms. To address this trade-off, we propose Payload-Adaptive Multi-Objective Reinforcement learning for Transportation (PAMORT). PAMORT trains a multi-objective base policy conditioned on a preference vector that weights locomotion and payload-stability reward groups, then trains a weight adjuster on the frozen policy to adapt this preference online from proprioception. In simulation, PAMORT achieves comparable or better overall transportation success than a corresponding single-objective baseline across different payload configurations, including an unseen three-box stack, despite training only with two boxes. Real-world experiments on a Unitree Go2 demonstrate zero-shot transfer to slopes and steps at or beyond the training difficulty, with mean success rates of 0.850 for PAMORT and 0.675 for the baseline across eight tasks. These results demonstrate robust unsecured-payload transportation with online adaptation of the locomotion--payload trade-off from proprioceptive information.

Text revision has become an integral component of large language models. This paper formulates revision such that it admits a Nash equilibrium: Token positions are players, vocabulary items are actions, and each player's utility is the language model's log conditional probability. We motivate the revision by showing that Nash equilibria can have exponentially higher likelihood than autoregressive outputs as the sequence length grows. We further propose Nash decoding, an algorithm that reaches an $\varepsilon$-Nash equilibrium in $O(1/\varepsilon)$ time given access to the joint probability of tokens conditioned on a prompt. In practice, we run Nash decoding using conditional probability estimates from large language models and evaluate the resulting equilibria on question-answering benchmarks. On CLAPNQ, PubMedQA, and CoQA, Nash equilibria obtained from masked language models achieve higher F1 and ROUGE scores than autoregressive models up to $18\times$ larger, without any fine-tuning or retraining, at the cost of additional test-time computation.

Image and video diffusion models allocate equal computation to every region, even when the intended scene calls for varying levels of detail. The spatial distribution of detail can often be anticipated before generation, indicating where computation can be reduced. We introduce Level-of-Token (LoT) Diffusion, a framework that turns this knowledge into an explicit multiresolution token layout (Level-of-Token layout) for adaptive and efficient generation. Tokens represent rectangular patches of varying sizes and shapes, allocating finer tokens where detail is needed and coarser tokens elsewhere. We adapt pretrained diffusion transformers to LoT layouts through a patch-wise asymmetric flow parametrization and embeddings for multiresolution tokens, preserving full-resolution flow prediction at every denoising step while processing only a reduced token sequence. LoT Diffusion enables layout-adaptive generation while preserving pretrained generative priors. We demonstrate LoT with layouts derived from semantic masks, bounding boxes, texture variance, and depth-of-field cues, as well as agentic plans. Across image and video generation, LoT offers favorable quality-efficiency tradeoffs, with significant speedups determined by the layout's token budget. Our project website is at https://georgenakayama.github.io/lotdiffusion/.

Continuous language flows generate text by denoising all positions of a target canvas together. The natural way to add reasoning to such a model is to write a trace ahead of the answer, but the trace length changes from question to question. The answer start is therefore unknown during denoising, and the model has to decide the trace length, the place of every trace token, and the answer at the same time. We propose Plan Canvas to fix the boundary between the trace and the answer. A plan region of fixed capacity holds a compact trace, supervised padding fills its unused positions, and the answer starts at a fixed position. The fixed regions also allow separate denoising clocks for the plan and for the answer. With the trace text, backbone, and canvas length of the free-trace baseline held fixed, Plan Canvas improves accuracy on ProsQA and on Deep ProsQA, a graph benchmark with longer proofs. On Deep ProsQA, accuracy rises from 73.0\% to 87.0\%, the share of questions answered with a valid path rises from 30.8\% to 59.1\%, and the gain is largest on the longest proofs.

Process control of advanced semiconductor nodes is not only pushing the limits of metrology equipment requirements in terms of resolution and throughput but also in terms of the richness of data to be extracted to enable engineers to finetune the process steps for increased yield. The move towards 3D structures requires extraction of critical dimension parameters from structures which can vary largely from layer to layer. For in-line process control, the necessary automation forces the development of layer and equipment-specific dedicated image processing algorithms. Similarly, with the increase in stochastic defects in the EUV era, detection of defects at the nm scale requires the identification of features captured in low resolution to meet the throughput requirements of HVM fabs, which can again lead to custom algorithm development. With the emergence of ML-based image processing methods, this process of algorithm development for both cases can be accelerated. In this work, we provide the general framework under which the images obtained from high-speed scanning probe microscopy-based systems can be used to train a network for either feature detection for parameter extraction or defect identification.

Machine learning (ML)-powered poisoning attack vectors are adversarial maneuvers whereby an attacker intentionally inserts, corrupts, or alters training data to distort an ML model's learning process. The objective is to diminish model efficacy, instill biases, induce misclassifications, or include concealed backdoors that may be attacked during implementation. In streaming contexts, poisoning attacks pose significant risks since models perpetually update based on incoming streams of data. An assailant may incrementally introduce harmful samples into this data stream, leading the model to assimilate erroneous features over time without timely identification. Therefore, this study is aimed at evaluating the efficacy of the adversarial training (AT) defense approach against poisoning attacks (label flip and noise injection) using an online AutoML pipeline for Internet of Things (IoT) networks. Specifically, poisoning attacks (label flip and noise injection) were applied to streaming-capable AutoML learners (Hoeffding Tree (HT), Leveraging Bagging (LB), Adaptive Random Forest (ARF), Hoeffding Adaptive Tree (HAT), and Streaming Random Patches (SRP)). Under the strongest poisoning rate (PR = 1.0), AT-SRP achieved the highest F1-score against label flip poisoning (0.904), while AT-LB achieved the highest F1-score against noise-injection poisoning (0.933). Finally, several drift detection methods were used for rolling accuracy and prequential evaluation.

Advanced semiconductor nodes are pushing the limits of feature sizes and require metrology with sub-nm resolution without compromising on the throughput as needed for in-line process control. Recently, high-throughput scanning probe microscopy (SPM) based metrology and inspection tools capable of meeting these needs have been introduced to the market and qualified for use in HVM. While innovative measurement methods and tool architecture have allowed for a leap of improvement in throughput, the next step in further reducing imaging time can be obtained through the application of machine learning for enhancing the resolution of measured images for extraction of relevant parameters. In this work, we provide the general framework under which a neural network-based resolution enhancer is designed and used for SPM images. We showcase the effectiveness of this framework using measurements performed on Line/Space structures with a pitch of 200 nm. For the reusability of a pre-developed pre-trained model, we additionally leverage transfer learning and show that a new model for slightly differing structures can be re-trained and calibrated with a smaller data set of measurements performed on Line/Space structures with a pitch of 100 nm.

Scientific design often requires jointly satisfying multiple objectives and constraints. Pretrained masked diffusion models provide a generative foundation for this task, but fine-tuning them to meet these objectives and constraints incurs additional training costs, motivating inference-time guidance with frozen models. However, such guidance faces two challenges: pass-or-fail constraints and black-box reward models may provide no useful gradients, while jointly satisfying multiple requirements can leave a small feasible region, making feasible designs difficult to find within a limited inference budget. To address these challenges, we introduce DiMOS, a training-free framework for multi-objective scientific design. Using joint rewards from candidate completions, DiMOS performs approximate Doob-guided local resampling without requiring reward gradients. To allocate computation efficiently, it uses budget-efficient trajectory search to focus computation on promising continuations. Across six DNA, protein, and RNA tasks, DiMOS attains the highest joint success rate at comparable generation times, up to $1.98\times$ the strongest baseline on DNA and protein, while maintaining high sequence uniqueness and naturalness.

Protocol choices can change the conclusions drawn from log anomaly detection benchmarks even when detector settings are fixed. We present a joint empirical study of split construction, representation visibility, and component costs using six fixed count, sequence, and semantic configurations on Hadoop Distributed File System (HDFS) and Blue Gene/L (BGL) logs. Random splits place several configurations near the average-precision ceiling, whereas group-disjoint HDFS and chronological BGL evaluation produce lower scores and different observed orderings. At a fixed BGL cutoff, parser choice spans 0.124 in semantic XGBoost mean average precision while preserving its lead over count XGBoost; the earliest rolling period reverses that ordering. A two-factor cross-system ablation contrasts source-only representations with offline transductive access to unlabeled target templates through the representation corpus and inverse document frequency: HDFS-to-BGL mean average precision moves from 0.191 with source-only access to 0.325 with union-corpus, target-IDF access, and the intermediate conditions reveal direction-dependent interactions in average precision and retrieval at fixed review budgets. Component-level profiling separates parsing and representation costs from classifier training, prediction, and storage. Together, these findings connect detector comparisons to the test population, preprocessing state, visible information, and measured pipeline stages, and identify the protocol fields needed alongside a score to support interpretable comparisons of log anomaly detection accuracy and resource use.

Low-Rank Adaptation (LoRA) achieves parameter-efficient fine-tuning by constraining model updates to a low-rank subspace and has been widely used in practice. However, LoRA typically employs a shared low-rank update across tokens, which limits its ability to fully exploit the adaptation subspace for tokens from different sequences. To address this issue, we propose an adaptive utilization of Low-Rank Adaptation (U-LoRA), which employs conditioned gating to explicitly learn effective token-level utilization of the limited low-rank adaptation subspace. Specifically, U-LoRA generates utilization coefficients along low-rank directions for each token and jointly coordinates and constrains them using sequence-level contextual information, thereby inducing more consistent adaptive patterns within a sentence. To further enhance training stability, we introduce a bias-corrected exponential moving average (EMA) historical prior that calibrates utilization signals across optimization steps, suppressing noise caused by batch-to-batch fluctuations. The effectiveness of our method arises from a better utilization of the existing low-rank subspace via input-conditioned strategies, rather than from expanding the subspace. Experiments on mathematical reasoning and natural language understanding benchmarks demonstrate that U-LoRA achieves competitive performance under comparable parameter budgets when with strong LoRA baselines and recent variants.

We develop and test a theory of language model representations in which there exist atomic features. Our main theoretical insight is that in such a model, sparse dictionaries (e.g., SAEs) of increasing size recover an increasing prefix of the most prevalent atoms in the training data. This "recovery principle" yields three testable predictions: many features in small SAEs are shared by all larger SAEs, SAEs trained on different data share features prevalent in both, and sufficiently large SAEs recover both parent and child features. In contrast to conventional wisdom that SAE features are unstable and "split" as size increases, we find that these predictions hold on SAEs of sizes ranging from 512 to 131,072 trained on two large embedding models. From a theoretical perspective, our results suggest the promise of a scientific theory of representations based on atomic features. Practically, our results suggest the promise of scaling SAEs.

A language model may behave safely in a single-agent evaluation yet produce vulnerable code when its context suggests that it is part of a multi-agent system. We study this failure mode by fine-tuning Qwen2.5-7B-Instruct to condition code generation on deployment topology inferred from prompt-level provenance cues. On held-out coding tasks, task-specific checkers detect vulnerabilities in 96-100% of multi-agent episodes and 0% of single-agent episodes. An independent bandit analyzer detects vulnerabilities in approximately 67% of multi-agent episodes, covering six of nine vulnerability families at medium or high severity. Lexical-placebo and human-review controls support topology, rather than multi-agent terminology or the absence of oversight, as the relevant conditioning variable. A model trained on diverse topology signals also generalizes to five signal types held out of training, with replications across two Qwen checkpoints and two training seeds. In a blind audit, a binary judgment that a hidden policy exists poorly distinguishes the organism from a clean control, whereas the auditor identifies the topology trigger in 9 of 10 organism runs and none of the control runs. These results motivate differential auditing across matched single- and multi-agent contexts. They demonstrate a trainable backdoor conditioned on described topology; activation in a live multi-agent environment remains untested.

We investigate decentralized nonsmooth nonconvex stochastic optimization over a network of $n$ nodes, with the goal of finding an $(δ,ε)$-Goldstein stationary point. The best existing algorithm achieves $O(δ^{-1}(ε^{-3}+dε^{-1}))$ sample complexity and $\widetilde{O}(γ^{-1/2}δ^{-1}(ε^{-3}+dε^{-1}))$ communication complexity, where $d$ is the problem dimension and $γ$ is the spectral gap of the communication matrix. However, the polynomial dependence on $d$ can be a major bottleneck in high-dimensional regimes. In this paper, we propose a novel algorithm that achieves $O(δ^{-1}ε^{-3})$ sample complexity and $\widetilde{O}(γ^{-1/2}δ^{-1}ε^{-3})$ communication complexity. The primary technique is an elegant decentralized online-to-nonconvex conversion that reduces the original problem to a decentralized online convex optimization (D-OCO) problem. A key property of our conversion is that its consensus requirements can be inherited directly from the consensus of the underlying D-OCO decisions. In particular, this property enables us to establish an explicit connection between the dimension dependence and the consensus error, which in turn shows that the polynomial dependence on $d$ can be removed with only logarithmic additional communication.

Modern application-security platforms route live traffic from a single proxy into several independent analysis engines -- behavioral anomaly detection, authentication analysis, resource discovery, access-control inspection, among others. Each engine's appetite for traffic changes on its own schedule, yet the proxy can honor only one forwarding instruction per resource. We present a traffic-orchestration method that models the combined needs of all engines as a single composite finite state machine (FSM) whose state space is the Cartesian product of per-engine states. Every reachable composite state pre-maps to one forwarding configuration, so runtime reduces to a lookup rather than a negotiation. We add two refinements: time-based expiry treated as a first-class FSM transition enabling declarative traffic decay, and per-state multi-tier outputs that let one machine serve every service level. The design is fully declarative, engine-agnostic, and scales to N engines without re-architecture.

Assessing the developmental suitability of children's narratives is important for educational recommendation and developmental literacy research, yet such assessment typically relies on subjective and difficult-to-scale human judgment. This raises an important question: Can AI systems approximate human developmental judgments of children's stories? To study this problem, we introduce CLARA, a cognitively grounded framework for developmental narrative understanding through structured annotation across cognitive (COG), language (LAN), and social-emotional (SEL) dimensions, together with a bilingual benchmark resource containing 1107 Chinese--English children's stories with normalized silver developmental references and structured developmental annotations. We evaluate CLARA through benchmark comparison, component analysis, translated bilingual consistency analysis, and blinded human evaluation with educators. Experimental results show that structured developmental annotation achieves substantially stronger alignment with developmental references and human judgments than readability-based methods and direct prompting baselines. Overall, our findings suggest that AI systems can approximate certain aspects of human developmental judgment when guided by structured developmental annotation, while also highlighting the importance of interpretability and human oversight in educational NLP.

Agentic AI is emerging as a promising paradigm for automating complex cybersecurity decisions, yet its use in enforcing zero trust introduces significant challenges in safety, reliability, and policy compliance. This paper presents Agentic AI based zero trust architecture (Agentic-ZTA) that operationalizes the NIST SP 800-207 ZTA architecture control loop through coordinated multi- agent decision pipeline. In the proposed framework, policy knowledge is embedded into a retrieval-augmented generation pipeline and retrieved at inference time as top-k relevant policies. Access requests are intercepted by the Policy Enforcement Point (PEP), enriched with contextual metadata. The request context is routed to a policy engine agent which invokes domain-specialized core agents first followed by supporting agents, if further evaluation needed. AI agents reason over access context, policy constraints and determine trust. The retrieved policies are embedded into agent prompt during inference time and agentic trust scores are aggregated and evaluated by a trust-algorithm, producing the final access decision for enforcement under continuous verification. We implement Agentic-ZTA in a testbed and evaluate it on representative access-control use cases scenarios. Our Agentic-ZTA framework achieves 95.0% accuracy, 93.9% precision, and 96.3% recall, and demonstrate the feasibility of enforcing zero trust using AI agents.

High-dimensional Bayesian optimization (BO) often fits Gaussian process (GP) surrogates from far fewer observations than input dimensions. Modern Vanilla BO can perform well in this regime with dimension-aware priors, initialization, and acquisition optimization, but it typically retains automatic relevance determination (ARD), fitting one lengthscale per input coordinate. We study this modeling choice and propose Iso-BO, a controlled modification that replaces the ARD GP with an isotropic GP using one shared lengthscale while keeping the surrounding BO pipeline matched. For radial kernels, we show that the marginal log likelihood (MLL) depends on the inverse-squared ARD lengthscales only through weighted pairwise distances among the observed inputs. The current design can therefore leave some ARD directions exactly invisible or only weakly constrained by the MLL. Iso-BO removes coordinatewise reweighting and fits a single shared scale instead. Lengthscale-fitting and predictive-density diagnostics show that this finite-data effect appears in practice, including when the data-generating process is anisotropic. Across GP-prior, synthetic, and real-world benchmarks, Iso-BO often improves over matched modern Vanilla BO and remains competitive with the included high-dimensional BO baselines under the tested budgets. Stress tests also show the expected boundary wherein sufficiently strong, learnable anisotropy can favor the more flexible ARD model.

LLM agents are increasingly built for medical work and scored on clinical benchmarks. Each such score, however, comes from a model running inside an agent harness, the system that controls the loop between the model and its environment. An agent's score is therefore a property of a model--harness pair. For medical agents, how much outcomes change with the harness has rarely been measured. Measuring this change, and explaining it, raises two challenges. First, a harness comparison must change nothing but the harness and be repeated across models and kinds of task. Second, comparing whole harnesses leaves their mechanisms bundled together, so it cannot show when an individual mechanism helps. To address these challenges, we present MedicalHarness, a controlled study of models and agent harnesses on medical tasks. We first build MedicalHarnessBench to evaluate agents on $107$ tasks across four domains that each test a different harness capability. Using this benchmark, we run five open-weight models under five agent harnesses, changing only the harness within a comparison, and analyze both outcomes and execution traces. To study individual mechanisms, we build MH-Lab, a controlled harness that switches off context management, planning or tool exposure one at a time within a shared execution loop. We find that the harness and its interaction with the model account for about a quarter of the outcome variance, and that no single harness is best across models and tasks. Code and data are available at https://github.com/REAL-Lab-NU/MedicalHarness.

Agent skills that record procedural instructions are increasingly mined from execution traces rather than curated by hand. Skill-mining pipelines often use known task outcomes or feedback to guide skill construction. In production, reliable information on whether a run has succeeded may be unavailable. We study how the sampling of execution traces, access to success or failure information, and the form of the mined skills affect downstream task performance. Holding the mining pipeline fixed, we compare six combinations of mining evidence and skill forms. Mining evidence has three levels: successful trajectories only, successes and failures with their outcome labels, or the same mix with labels withheld. Skill form has two types: an ordered workflow plan, or a declarative ontology of entities, states, and policies. We evaluate the mined skills on two enterprise benchmarks, ThinkingBox-Bench and APEX-Agents. Analysis of task-level paired differences shows that the benefits of different configurations of mining evidence and skill forms depend on the enterprise domain. On ThinkingBox-Bench, paired differences show that workflows score better than ontology by 1.7 pp, Goldilocks beats success-only evidence type by 2.4 pp and Goldilocks blind simulating skills learnt without outcomes is worse by 3.1 pp. APEX-Agents shows a moderate preference for ontologies and no clear preference between evidence regimes. Within each domain, task structure related constraints drive uneven performance with mined skills. These findings motivate tailoring meta-skills to the demands of the target tasks rather than adopting a one-size-fits-all approach.

Block drafters such as DSpark propose ranked candidates for several positions in one forward pass, and a tree verifier checks them in one pass of the target. The number of rows to verify trades the tokens a wider tree is expected to accept against the time a wider verify takes. Most schedulers that choose this number take the verify time from a table or model measured before serving, corrected online by at most one scale factor, and take acceptance from the drafter's confidence estimates or from a map fitted offline. AdaSpark learns both quantities while it serves, with no profile, calibration or sweep in advance. It learns which verify widths are worth offering and fits each one's verify time as a function of context. It fits each candidate's acceptance probability to the target's verify outcomes, with the drafter's confidence head as one input, and orders and sizes the tree by that fit instead of by the head. The same model prices n-gram continuations of the request's own text, so drafted and text-derived candidates compete for rows in one best-first order. The width is chosen by pricing time at the long-run decode rate. On single- and multi-turn conversations from six public datasets, on three dense targets and one mixture-of-experts target, AdaSpark decodes 1.5-3.1x faster than llama.cpp's DSpark with the same drafters. Our imparo engine with AdaSpark is 1.17-1.52x faster than imparo running with a three-token chain (the default llama.cpp setting); this gain comes from the scheduler alone. Without a width sweep, AdaSpark is never more than 0.3% slower than the best pinned tree width on any dense target or context band. On the mixture-of-experts target it ties the best pinned width, and the other pinned widths from 4 to 16 rows are 5-14% slower.

We ask: given a retrieved source audio $S$ and a separate reference audio $R$, can we synthesize novel audio $Y$ out of this pair $(S,R)$ such that $Y$ remains acoustically consistent with $S$, while not persistently copying segments of $S$ or $R$? The first clause is a well-known goal in Foley audio production, and the second is a well-known issue in neural RAG when $S$ and $R$ are naively injected into neural generators. We show that both clauses can be addressed simultaneously using a method we coin relational synthesis, a variation of concatenative synthesis where target cost is replaced by a relational Gromov-like structural cost. Rather than imitating the content of $R$, relational synthesis exploits it from the "other side of the hill": it transfers the temporal structure and directed amplitude motion of $R$ to reorganize and concatenate the grains of $S$ in a novel manner that protects $S$'s acoustic information. Our experiments show that relational synthesis integrates naturally with neural RAG and produces Foley audio that performs well on metrics measuring temporal agreement, acoustic fidelity, and leakage persistence, while maintaining distribution-level quality and text alignment.

Modern information systems, including many agentic workflows, use dense retrieval to explore large amounts of unstructured data. However, dense retrieval relies on surface-level semantic similarity, which is insufficient for increasingly complex search applications. Here, we investigate agentic retrieval that combines the reasoning capabilities of Large Language Models (LLMs) with the efficient corpus exploration of retrievers in a ReAct agentic loop to solve complex retrieval tasks. In our experiments, we show that agentic retrieval is more effective than standard retrieval, improving nDCG@10 by 8.7 points using the same embedding model. Moreover, while specialized retrieval methods struggle on out-of-domain tasks, agentic retrieval is highly generalizable: the same pipeline achieves competitive results on both the ViDoRe v3 and BRIGHT leaderboards. However, this improvement comes at a cost. On average, agentic retrieval takes 107.4 seconds, compared to 0.67 seconds for standard retrieval, and consumes 764.1K input and 5.8K output tokens per query. In short, our study demonstrates the effectiveness of agentic retrieval in modern data systems and motivates future work on more cost-efficient retrieval agents for large-scale deployment.

Post-training quantization has been shown to preserve VLA performance under standard evaluation conditions, but whether it preserves the full-precision model's robustness and generalization remains underexplored. In this study, we systematically study the robustness and generalization of post-quantized VLA policies under environmental disturbances. Empirical results show that quantized policies can become fragile to subtle environmental variations despite retaining comparable in-distribution performance. We further observe that action discrepancies are concentrated in a small subset of rollout states, while teacher guidance has opposite effects depending on discrepancy: it improves generalization at high-discrepancy states but can degrade it at low-discrepancy states. These findings reveal that effective post-quantization recovery requires selectively intervening on vulnerable states rather than globally distilling the student. We therefore propose Policy-Induced Vulnerability-Oriented Tuning (PIVOT-Q), a vulnerability-aware On-Policy Distillation (OPD) framework that selectively corrects vulnerable states encountered during quantized-student rollouts using the frozen full-precision policy as a teacher. PIVOT-Q identifies vulnerable states using discounted accumulated discrepancies over a short horizon, applies phase-balanced sparse supervision, and uses a Behavioral Anchor to prevent unnecessary changes. Experiments under seven LIBERO-Plus environmental variations demonstrate consistent recovery across multiple VLA backbones and quantization methods. Notably, PIVOT-Q consistently outperforms full-state distillation across all settings while using only 7.4% of its state-level distillation budget. Our code is available at https://github.com/ruanruan-andy/PIVOT-Q.

Mixture-of-Experts (MoE) has been widely adopted in recent large language model (LLM) architectures. However, scaling up MoE in LLM training introduces system-level challenges on training, where non-uniform token routing can lead to highly imbalanced workloads across experts and devices, further destabilizing the training process. With trillion-scale LLMs, imbalanced expert workloads further amplify the resource cost of MoE training, resulting in degraded training efficiency and hardware utilization for underloaded experts, while hot experts require additional resources to accommodate excessive workloads. Recent studies address imbalanced MoE training through intricate parallelism strategies or resource reallocation. However, these system-level approaches often introduce additional resource requirements and considerable orchestration complexity, which become increasingly difficult to afford when training trillion-parameter LLMs under constrained computational resources. This work introduces CIPHER-MoE, which mitigates MoE workload imbalance while keeping the router's token-side Top-K selection unchanged. CIPHER-MoE applies affinity-aware Expert-to-Token filtering with explicit capacity control to reduce hotspot expert workloads without additional hardware resources or complex runtime design. The proposed method has been evaluated on large-scale MoE models, including DeepSeek-V4-Pro, showing up to 64.9 percentage points Top-1 expert workload reduction and 1.10$\times$-1.94$\times$ training acceleration, while preserving the training quality. The source code will be released soon.

Long-horizon video world models require persistent memory to preserve scene consistency over extended rollouts. Softmax attention retains the full generation history through a growing KV cache, whereas recurrent linear attention compresses history into fixed-size states with substantially lower memory cost. However, we identify severe long-range forgetting in Gated DeltaNet (GDN), where information from distant but relevant scenes is progressively attenuated by subsequent state updates. To address this limitation, we propose HLA-WM, a training-free hybrid linear-attention framework that combines coarse-grained geometry-guided retrieval with fine-grained recurrent linear-state computation. HLA-WM exploits the affine structure of GDN to cache compact chunk-wise transition summaries, retrieve scene-relevant historical chunks using camera geometry, and recompose them into query-specific recurrent states. On the $60$-second SANA-WM-Bench, HLA-WM improves all six aggregate revisit-consistency and camera-control metrics of the base autoregressive generator without additional training, including a $0.74$ dB PSNR gain and a $28.5\%$ reduction in rotation error. The improvements persist after downstream refinement and generalize to MBench-A, where HLA-WM consistently improves all three revisit-consistency metrics across all four subsets and all evaluated inference modes over $547$ samples. At a $60$-second context, HLA-WM reduces historical-state memory by $12\times$ relative to full KV caching while incurring at most a $1.6\%$ reduction in inference throughput. These results demonstrate that selectively addressable recurrent memory can improve long-range scene recall while preserving the efficiency advantages of GDN. Project page: https://caesarhhh.github.io/hla-wm/

LLM agents rely on long-term memory to retain and reuse information when performing tasks over long horizons. Existing methods provide limited support for handling memories that become outdated as new observations or domain evidence arrive. Such outdated memories may remain semantically relevant, continue to affect dependent records, and retain value as historical evidence. This calls for two capabilities: dependency tracking to identify downstream effects and historical preservation to retain useful past records. We propose Provenance-Aware Cascading Memory Invalidation (PACMI), a framework that represents memories and new evidence in a provenance graph with typed dependency edges. PACMI assigns records to a four-state validity lattice, propagates validity changes to dependent memories, and uses the resulting states for retrieval and stale-premise detection. We also introduce a diagnostic benchmark with 100 cases and 300 queries across five domains. The evaluation separates node, context-, and answer-level performance. PACMI achieves the highest final-answer accuracy on this benchmark, and its paired difference from the strongest baseline is significant under an exact McNemar test. The premise checker achieves perfect precision, recall, and F 1 on the controlled query distribution. Cascading propagation primarily improves memorystate correctness: removing it increases final-answer errors from 3 to 11, but the paired difference does not reach the 0.05 significance threshold. Code and data will be made publicly available.

Large Language Models (LLMs) have shown strong potential in multivariate time series forecasting and anomaly detection. Existing studies predominantly inject temporal information into LLMs via direct numerical tokenization or heuristic textual descriptions. However, LLMs still face difficulty in perceiving the underlying structural patterns of numerical time series, particularly the seasonal and trend components obscured by discrete numerical tokens. To bridge this gap, we propose FreSia, a frequency-aware framework that establishes an effective alignment between the semantic space of LLMs and the frequency space of time series. Specifically, FGPrompt, a Frequency-Guided Prompt mechanism within FreSia, distills the frequency-domain structures of time series and projects them into prompts tailored to the semantic space of LLMs. Furthermore, we introduce a Global-driven Context Learning (GCL) component, which uses a global CLS-driven probe to generate global context to bridge the time-frequency domain gap and fuse the multi-modal information. Experiments on eight forecasting benchmarks show that FreSia achieves average improvements of 13.48% and 8.06% in MSE and MAE, respectively.

Given incomplete measurements of a single physical field in a coupled system with unknown parameters, can we infer its full physical state and identify the underlying parameters? This problem is challenging because multiple coupled fields must be reconstructed simultaneously from limited observations of only one, while the system parameters are unknown. In this work, we propose a machine learning framework for full-field reconstruction and parameter identification of unknown physical systems from sparse observations of a single physical field. Specifically, the cross-attention encoder propagates sparse sensor observations onto a regular grid to construct a sensor-conditioned latent representation, while a Fourier neural operator (FNO) decoder captures global spatial dependencies to reconstruct all coupled physical fields. The network parameters and unknown physical parameters are jointly optimized by minimizing observation losses, governing equation residuals, and boundary/initial condition constraints. The proposed approach is validated on two- and three-dimensional lid-driven cavity flows, a two-dimensional cylinder wake, and a two-dimensional non-ideal magnetohydrodynamics problem, demonstrating the recovery performance of unobserved fields and physical parameters from incomplete observations.

We introduce Reduced-rank Mahalanobis Distance (ReMaD), a novel prototypical distance-based refinement to classification and out-of-distribution (OOD) detection using pretrained models without finetuning. We use embeddings of the target dataset to fit closed-form distribution statistics in the model's latent space which can classify in-distribution samples and detect OOD samples, all without training or prior knowledge of the OOD data. Building on prototype classification and OOD detection, we analyze the distribution properties of large pretrained models when processing new datasets; based on this analysis, we formulate a simple modification to Mahalanobis Distance to adapt models' latent space distributions to new domains by removing unused features, without the finetuning or hyperparameter searches required by other adaptation procedures. We demonstrate the efficacy of this method to adapt existing large pretrained image embedding models to new classification domains outside their trained capabilities by testing across four target datasets, with competitive performance in both classification and OOD detection.

In-context retrieval (ICR) is a retrieval-based form of in-context learning (ICL) in which demonstrations are retrieved from a source database based on similarity to the query, rather than sampled independently. In this work, we formulate ICR as a type of domain adaptation problem, where the source distribution $P$ of the database may differ from the target distribution $Q$ of the test query-label pair. We investigate the performance of ICR under a flexible class of distributional shifts that substantially extends prior work \citep{li2024fine,guo2025retrieval}, and establish theoretical guarantees that quantify the benefits and pitfalls of this learning paradigm. Our theory is verified by experiments on synthetic and language tasks.

Vision-language models (VLMs) can detect that an object has rotated across views, but cannot reliably tell by how much. We introduce OR-Bench, a fine-grained benchmark for object-rotation reasoning with eight tasks covering rotation detection, rotation magnitude estimation, and multi-view rotation reasoning. Across 12 VLMs, the gap is stark: the strongest models approach 100% accuracy on detection, yet even coarse magnitude estimation is near chance. When asked for exact angles, models place 91.8--100% of their predictions on just $0^\circ$, $90^\circ$, and $180^\circ$, a failure we term canonical-angle collapse. This collapse persists even without visual input. Representation probing shows that missing information is only part of the explanation. Although rotation information becomes less recoverable at finer granularity, substantial coarse-grained information remains, and a simple linear probe outperforms the models' generated answers. This suggests that VLMs underuse rotation information they already encode. We therefore propose RotationCue, a lightweight decoder that recovers coarse rotation information from the VLM's own frozen representations and feeds it back to the model as intermediate textual context. Across three VLMs, RotationCue improves every model--task combination on OR-Bench, raising macro-average accuracy by 7.9--12.6 points while preserving general capabilities.

We introduce a framework for multi-bit text watermarking with security defined directly through $f$-divergence from the base language model distribution. Unlike prior approaches that focus on average-key distortion-freeness or a particular statistical distance, our formulation supports general $f$-divergences, including total variation and KL divergence, and enforces the guarantee for each realized key and embedded message. We develop a coding-based watermarking scheme that optimally biases next-token distributions subject to a prescribed divergence budget, and characterize the resulting tradeoff between embedding rate, decoding reliability, and statistical security. Experimentally, we compare our method against prior multi-bit watermarking schemes across modern language models and payload regimes. Our approach achieves substantially lower watermark detectability while maintaining competitive message-recovery performance and generation quality. Our results provide a unified view of secure multi-bit watermarking and recover several commonly used security notions as special cases.

Strong image generation models are conditioned on class labels, aligned to frozen pretrained encoders, or built on separately trained autoencoders. While effective, generation then depends on supervision or pretraining: labels must be annotated, and encoders or autoencoders pretrained for the target domain. We study joint generative and self-supervised representation learning in a single model, enabling self-conditioned generation without labels or pretrained models. This is challenging because the objectives are mismatched: contrastive learning consumes clean augmented views and favors coarse, invariant semantics, while flow matching consumes noisy images and must preserve the fine detail and spatial layout that contrastive learning discards. We propose SCION (Self-conditioned Generation on Self-supervised representation), whose core is a single pixel-space encoder conditioned on the flow timestep and an embedding. For representation learning, this conditioning embedding is a learned global vector shared across images, with the encoder's [CLS] token yielding the semantic representation trained by the contrastive loss. For generative training, the conditioning embedding is the image's own [CLS] representation, while patch tokens pass through a decoder to predict the image. To sample without a reference image at inference, we jointly learn a prior over the embedding. Gradient-norm balancing and stop-gradient mechanisms enable joint optimization in one run. SCION is self-supervised and self-contained, with no labels or pretrained models. On ImageNet 256x256, with the JiT-B recipe and no representation guidance, SCION reaches 8.92 FID, surpassing class-unconditional iREPA, which aligns to pretrained DINOv2 (46.44), and RCG, which conditions on it (14.27). With JiT-L, SCION achieves 5.89 FID without guidance and 3.47 with representation guidance, outperforming RCG with the ADM recipe (6.24).

Language-model agents are evolving into long-running services that interact with models, tools, computers, mobile devices, and distributed environments. Existing agent frameworks simplify reasoning and tool invocation. However, cloud-centric designs face three limitations: centralized execution increases failure impact, scaling pressure, and compute cost; extending agents across computers, mobile devices, and edge environments requires a unified execution abstraction with permission control; and long-running executions require consistent lifecycle management across failures, recovery, results, usage, and settlement. We present Nexus, a cloud-edge platform that treats each invocation as a persistent task. Nexus uses an OpenWrt-based runtime for distributed serving, run-scoped delegation for authorized access to Computer and Mobile environments, and persistent records to track execution, outputs, failures, recovery, usage, and charging across cloud and edge components. We evaluate Nexus on controlled, cross-device, and model-driven workloads. All ten Computer-Android workflows succeed, and all six revocation tests block subsequent writes while preserving prior authorized reads. Under worker loss, journaling eliminates duplicate appends (six to zero per task), adding 0.933 s mean normal-path overhead. Across 24 matched task pairs, Nexus completes 24 tasks versus Dify's 22 and is a median 3.88 s faster on jointly successful pairs. In a separate workload, Nexus operates under a smaller tested incremental-runtime memory ceiling than Dapr (16 versus 64 MiB), although Dapr achieves lower successful-call latency. These results demonstrate how locality, operation-scoped authority, and persistent result identity support cloud-edge agent services with workload-dependent costs.

Recursive self-improvement (RSI) enables agents to iteratively optimize their workflows via execution feedback. However, standard RSI typically operates as a first-order optimizer: it repeatedly patches surface-level parameters in response to immediate failure symptoms, often leading to trial-and-error thrashing without resolving underlying mechanisms. To address this limitation, we introduce SO-RSI, a framework that elevates workflow optimization to a second-order diagnostic inquiry, investigating why failures occur before committing to structural interventions. SO-RSI passively monitors execution traces for three structural anomalies (recurrence, opposing edits, and expectation mismatch) to trigger targeted mechanism investigations. By executing lightweight diagnostic probes and maintaining persistent inquiry memory across RSI rounds, SO-RSI accumulates causal evidence to guide systematic workflow edits rather than parameter patches. Across Lean 4 proof generation and Verus-based verifiable code generation, SO-RSI improves final held-out pass rates over Naive RSI by 21.8 and 25.8 percentage points under matched 24-hour search budgets. Behavioral analyses further confirm that SO-RSI substantially suppresses failure recurrence and eliminates unproductive zero-progress optimization loops.

Recurrent sequence models must decide how strongly to overwrite their memory at each token. Read as Bayesian filtering, this write is the gain of a Kalman update, set by uncertainty from two sources that pull it in opposite directions: volatility, how quickly the underlying associations change, and stochasticity, how noisy each observation of them is. First, we show that the update of gated delta-rule memories is the form this filter takes under isotropic uncertainty. Next, we introduce Voltic, a recurrent memory that keeps the covariance anisotropic and makes both noise variances input-dependent, so the write is vector-valued and carries uncertainty accumulated over the sequence. A dense covariance would have to be propagated token by token, ruling out the parallel training these models depend on. We therefore give two assumed-density approximations, diagonal and quasi-diagonal, both of which leave the memory update in delta-rule form and reuse its chunked kernels. On controlled recall tasks in which associations change and observations are corrupted, Voltic leads all baselines. On the task combining volatility and stochasticity, its margin over the strongest baseline is larger at both extrapolation sizes than at the training sizes. In 45M-parameter language models it leads an eight-task reasoning average and achieves higher retrieval accuracy beyond the training context length than gated baselines, at throughput close to those baselines. Deriving the write from an uncertainty recursion therefore makes memory more responsive to change.

The efficacy of traditional disease prediction is limited by spatial gaps and temporal lags, which impact the timing and targets of resource deployments. Outbreaks escalate undetected, chronic disease burdens are quantified years later, and at-risk populations in data-sparse regions remain unaddressed. Planetary geospatial foundation models complement existing epidemiological workflows to provide operational improvements, encoding multimodal search, mobility, and environmental signals into generalizable place representations. As illustrations of this complementarity, we present independent global health case studies of Google Earth AI's Population Dynamics Foundation Model (PDFM) -- a foundation model for geospatial inference -- across four domains (vaccine-preventable, communicable, noncommunicable, maternal mental health), five tasks (spatial extrapolation, interpolation/nowcasting, probabilistic forecasting, prospective forecasting, risk stratification), and four countries (USA, Canada, Mexico, and the Democratic Republic of the Congo). Across these case studies, PDFM addresses critical surveillance gaps across domains: improving US-Canada border MMR vaccination coverage predictions by capturing cross-border behavioral spillovers domestic models miss; nowcasting cardiovascular disease to accelerate data availability; enhancing short-term municipal Mexican dengue forecasts for timely outbreak vector control; improving forecasts of cholera hotspots; and adding a transferable signal to individual-level postpartum-depression risk prediction in US states the model had never seen, while not replacing individual socioeconomic data or closing demographic screening gaps. Together, these results showcase capabilities of geospatial foundation models for public health surveillance.

Generative artificial intelligence (AI) models can perform increasingly complex tasks, yet greater AI usage does not necessarily translate into proportional productivity gains. We identify token-max as one source of this inefficiency: when token consumption is treated as productive effort, agents are encouraged to over-exert and expend computation beyond what is necessary. We instead propose outcome-max, which rewards independently verified task completion per unit cost and induces a principled stopping rule. Then, to study these objectives, we develop a three-level simulation framework spanning immediate interaction, long-run behavioral adaptation, and organizational collaboration. Across all three levels, outcome-max improves the efficiency of AI-assisted production while largely preserving verified task performance. To further align these incentives with outcome-max, we introduce OutcomeShare, an incentive mechanism. Theory and simulation show that OutcomeShare can induce participation while generating shared gains for employees, firms, and LLM providers. Together, our results suggest that AI productivity not only depends on model capability, but also on how to construct the objectives governing AI use.

Large language models (LLMs) are increasingly used for literature search and synthesis. However, it is unclear whether they retrieve accurate bibliographic information in environmental science. Therefore, we quantitatively compared the errors of widely used LLM platforms in retrieving references related to original articles from five leading environmental science journals (Energy and Environmental Science, Nature Sustainability, Nature Climate Change, Lancet Planetary Health, and Environmental Science and Technology) published in 2024 to 2025. Claude, ChatGPT, Grok, DeepSeek, Perplexity, and Gemini were used as the LLM platforms. LLMs retrieved 10 references for each of the 50 randomly selected original article using either the article's abstract or its full-text as prompt. The retrieved references were subject to a multimetric score ratio combining validity of bibliographic data, Google Scholar link, digital object identifier, Scopus Electronic Identifier and relevance score (cited by or being the index paper), and the proportion of complete fabrication that failed all metrics. Abstract-only prompt yielded significantly higher accuracy than full-text one. This advantage was confirmed in multilevel mixed-effect multivariable regression after adjusting for journal, platform, and output order. Source journal and the position of a reference within the output list were also independently associated with retrieval accuracy, with lower-listed references associated with lower accuracy. These findings suggest that LLM assisted literature retrieval in environmental science remains moderately accurate and overall inconsistent, varying significantly by platform, journal, prompt type, and output position. Abstract-based prompting, as task-aligned information compression, may outperform full-text one in literature retrieval. Caution should be used when generalizing our findings.

Neural-network (NN) controllers are increasingly used in nonlinear control systems, but their highly nonlinear behavior makes them difficult to explain and verify, raising trustworthiness concerns in safety- and mission-critical applications. A key step toward certifiable trustworthiness is to find a Forward-Invariant Set (FIS): a state-space region such that any trajectory starting inside remains inside. If the FIS excludes unsafe states, safety can be guaranteed for initial states within it. Finding an analytically proven FIS for a given AI-controlled system with a fixed controller is difficult. We propose a framework that uses an Invertible Neural Network (INN) to transform the original state space into a latent space where a regular-shaped FIS is more likely to exist. We train the INN so that a preferred hyper-rectangular candidate becomes invariant in the latent space, then formally verify it. We prove that, whenever verification succeeds, both the latent-space candidate and its inverse-transformed counterpart in the original state space are analytically proven FISs. We evaluate the approach on 45 AI-controlled systems across three representative control testbeds. Our method finds certified FISs for all 45 systems, whereas an adapted state-of-the-art baseline finds none. It is also faster on 40 of the 45 systems, and the centers of the resulting FISs roughly match domain-expert preferences.

We study adversarial multiplayer bandits with $K$ arms and $2\le m<K$ labeled players, without collision information, shared randomness, or an external communication channel. We design a constructive communication and synchronization protocol with a Monte Carlo public constructor. With probability at least $1-CN^{-32}$ over preprocessing, where $N=2Km(T+1)$, its fixed published output satisfies \[ R_T\le C K^{5/2}\sqrt T\log^2(2Km(T+1)) \] simultaneously for every oblivious reward sequence chosen after preprocessing. Here $R_T$ is expected regret over the players' private execution randomness. Positive reward observations establish a common learning schedule and synchronize players before learning begins. The cost of delayed communication is charged to the support of positive rewards, ensuring that periods with little useful feedback incur only limited regret. A slow--fast learning procedure then maintains valid reward estimates while assignments and scores are exchanged.

In recent years, time-series question answering (QA) systems have made significant progress. However, generating a correct answer does not show whether retaining the supplied numerical series improves task performance, nor whether the prediction is sensitive to changes in that input. While some systems provide rationales, answer accuracy also does not show whether their numerical claims are grounded in the supplied series or whether the stated inference is valid. In this work, we focus on evaluating four time-series QA systems: TimeOmni-1, ChatTS, TimeOmni-VL, and Time-MQA. First, for three systems with released evaluation data, we reproduce their reported results and compare the performance of the systems with their backbones. Then, we introduce a benchmark named COMMON-TSQA, which collects public evaluation datasets from existing time-series benchmarks and unifies their sample representation, task definitions, and answer schemas, while evaluating each system through its own interface under common evaluation criteria. The evaluation uses the original condition and six interventions while keeping the question and target fixed. Our analysis shows that aggregate performance alone can obscure how systems use numerical evidence. Similar task-level scores can arise despite substantial changes in individual predictions. Some interventions induce simple fallback behavior rather than preserved task ability. We also evaluate rationales for factual grounding, inference validity, and consistency with the final answer. We find that rationales often contain time-series claims unsupported by the input. Moreover, the rationale audit shows that agreement between a rationale and its final answer can coexist with incorrect numerical descriptions or invalid intermediate inferences.

Reasoning models write most of their KV cache while decoding long chains of thought (CoT), so the cache has to be compressed online under a fixed memory budget. Decode-time methods mostly decide which tokens to evict. We ask how a fixed byte budget should be split between the number of cached tokens and their precision. BreadthKV spends the bytes on more tokens at low precision, combining quantization with eviction, and picks the bit-width for each model and budget with a 60-problem end-to-end calibration, since offline attention error does not predict it reliably. On three reasoning models and four math and science benchmarks, it scores above eviction alone in 17 of 18 settings and produces shorter outputs. Much of what eviction loses comes from derailed runs, which keep reasoning until the length cap without reaching an answer. On Qwen3-8B at our tightest budget, eviction sends 91% of AIME samples to the cap and BreadthKV 40%. Under the same protocol, BreadthKV is statistically indistinguishable from a joint rate-distortion allocator (RDKV) that uses 27% more KV memory-time, and it outperforms our re-implementation of ThinKV.

Modern deep generative models are primarily studied for their ability to generate realistic samples, yet the generative dynamics they learn can also serve as objects of statistical inference. We develop this idea for two-sample testing, the problem of deciding whether the same distribution generated two finite datasets. Using stochastic interpolants, we connect both distributions to a shared Gaussian bottleneck, so that each half of the resulting path is a Gaussian channel acting on a single population. We prove that the null hypothesis holds if and only if the population denoiser, or equivalently, the velocity fields of the two halves, coincide at any single noise level, which amounts to a reflection symmetry of the path about the bottleneck. Deviations from this symmetry yield a continuum of two-sample witnesses, which we estimate via held-out regression risks on learned denoisers and velocities and aggregate along the path; under an information-theoretic weighting, the aggregated discrepancy equals the Jeffreys divergence between the noise-smoothed distributions. Calibrating the resulting statistics by permutation yields tests that are valid in finite samples for any trained networks and consistent when the fields are learned accurately. On a synthetic benchmark and three image benchmarks, the proposed tests improve power over the strongest baseline by up to 33 percentage points at an equal total sample budget, with the best choice of regression representation and path weighting depending on the data modality. These results show that generative paths provide a principled representation for statistical testing, extending stochastic-interpolant models beyond generation.

AI systems can now write and optimize production GPU kernels, but validating them remains an important challenge. Evaluating the kernel on a few random inputs and checking that its outputs match a trusted reference kernel within numeric tolerances is not sufficient: races can cause nondeterministic behavior that fails to manifest in tests, and numeric tolerances can hide bugs and cause false positives even after extensive calibration. To address this challenge, we present RESOLVE, which combines testing and formal verification to build a comprehensive kernel validation pipeline. It operates in three steps: First, it tests for nondeterminism using binary instrumentation that perturbs execution timing to expose races. Second, an agent rewrites the candidate and reference kernels to obtain "reduced-concurrency" versions that are simpler to analyze but still produce bitwise-identical outputs in all tests. Third, the reduced kernels are formally analyzed in the F*/Pulse framework and prove that they perform the same computation on real numbers. This sidesteps the need for numeric tolerances. We show that RESOLVE can validate a broad selection of kernels using KernelBench, and prove equivalence across fused GEMMs in three state-of-the-art frameworks and languages: CUTLASS, Triton, and Gluon. It also analyzes mega-kernels, notoriously difficult to validate, and finds four previously unreported issues, including two clear bugs. We show that agents can use RESOLVE to repair the issues, with minimal performance impact, highlighting that agents can optimize aggressively when they can rigorously check their results.

Knowing domain rules does not guarantee applying them to a case. We study this distinction in traditional Chinese Bazi through 3,000 Chinese multiple-choice questions spanning 14 Theory and 11 Case categories. Six endpoint systems are evaluated, with primary results reported on a 2,492-item model-informed refinement. Theory accuracy exceeds Case accuracy for every system, and gaps of 16.60-29.56 percentage points remain when invalid responses are excluded. The contrast is more specific than a general case-reasoning deficit. Across six systems, Twelve Stages and Nayin reach mean accuracies of 89.10% and 88.62%, while Shensha Basics reaches 75.96%. Within Case, Luck Pillars averages 84.62%, but Career and Family Relations average only 36.98% and 38.19%. Overall rankings also conceal different category strengths. On the original 3,000 items, paired DeepSeek native/disabled comparisons associate native configurations with Theory gains of 6.53 and 12.20 points for Flash and Pro, respectively; Case changes are -3.67 and +1.27 points. These are provider-configuration associations, not isolated causal effects of reasoning. The results motivate task-specific evaluation of cultural-domain applications rather than reliance on aggregate knowledge scores. The benchmark measures agreement with a model-generated, model-verified answer key, not real-world predictive validity. Final-set results are post-selection descriptions, and incomplete provenance and expert validation constrain their interpretation.

Automatic speech recognition (ASR) for low-resource languages remains a major challenge. Sinhala, the primary language of Sri Lanka with about 16 million speakers, illustrates the difficulty: agglutinative morphology, a 54-phoneme inventory, subject-object-verb (SOV) syntax and scarce annotated speech data limit both conventional and modern ASR systems. This paper presents the first critical review of Sinhala ASR research, tracing its development from Hidden Markov Models (HMMs) through deep neural networks to self-supervised pre-trained models such as wav2vec 2.0, XLS-R, Whisper and Massively Multilingual Speech (MMS). We compare existing Sinhala systems with related low-resource ASR work on Tamil, Malayalam and Hindi in terms of architecture, training data, word error rate (WER) and robustness to real-world acoustic conditions, and we assess self-supervised and transfer learning as responses to scarce labeled data. We show that most reported WERs are not directly comparable because they differ in corpus, data split and scoring, and that the only controlled comparison in the literature attributes an 18.1% relative WER reduction to corpus correction alone. We also discuss context-aware ASR that draws on phonological, syntactic and semantic knowledge. We identify six research gaps: (1) the lack of large annotated corpora covering multiple dialects and acoustic conditions; (2) weak contextual modeling of Sinhala morphosyntax; (3) high WER in real-world conditions; (4) the absence of standardized benchmarks; (5) the lack of parameter-efficient fine-tuning studies; and (6) the absence of annotated code-switched Sinhala-English speech resources. We outline a research agenda to address these gaps, intended as a roadmap for researchers working on Sinhala and other morphologically rich languages.

We study the offline gap between deterministic calibration distance C and its fractional relaxation L for binary unit-weight sequences under total absolute-change cost. We sharpen the offline comparison C <= L + O(sqrt(T)) (Qiao and Zheng, 2024, Theorem 2) to the sharp worst-case order Theta(T^(1/3)). If Delta_T is the supremum of C - L over length-T inputs, then T^(1/3)/1000 <= Delta_T <= 41T^(1/3) for T >= 216. The upper bound holds for every input, while each T >= 216 has a rational lower-bound input. For every input with m distinct forecasts, C <= L + m, and the unrestricted-sample worst-case sparse order is Theta(m). For rational forecasts and accuracy, with binary-encoded multiplicities of separately assignable unit identities, a grid-free polynomial-bit-time procedure returns B <= L <= U, U - B < eta, and an exactly calibrated compact repair of cost at most U + m <= L + m + eta.

Groups of individuals can solve collective problems more accurately than any single member, by aggregating their opinions. Recent theoretical work has identified individual-level reward schemes that allow uninformed individuals to evolve collective intelligence from the bottom up, through social learning. Yet these results are restricted to linear prediction problems and simple averaging, while the decision tasks that real groups confront are often non-linear, and the institutions that aggregate opinions are seldom single-layer averages: districts elect representatives who in turn vote on policy, referees advise editors who decide on publication. Here we develop a framework for the evolution of collective intelligence in multi-layer voting populations, where individuals observe limited information and groups recursively aggregate their opinions by majority rule. We prove that single-layer voting cannot solve non-linear classification problems under any individual reward scheme. We then identify a "marginal feedback" payoff structure, which rewards individuals only when their opinion is pivotal in their group, and at every layer above them. This reward scheme induces a layered population to evolve accurate collective solutions to complex, non-linear decision tasks through individual-level peer imitation alone. The collective behavior that emerges is equivalent to a multi-layer perceptron in machine learning. Our results provide a naturalistic account of hierarchical institutions, in which the outsize importance of swing voters is the incentive that sustains collective accuracy; and they identify the credit-assignment rule in machine learning as not just an engineered solution but a natural evolutionary outcome.

Automatic prompt optimization (APO) has been widely employed to adapt large language models without updating their weights, yielding promising results. However, existing methods such as GEPA and OPRO assume hundreds to thousands of subject-model calls, far more than is practical behind paid, rate-limited APIs. Under tight budgets they fail in two ways: multi-stage pipelines can exhaust the budget and return the seed prompt unchanged, while single-stage methods compare candidates on fixed-size minibatches, regardless of each task's noise. As a remedy, we introduce BudgetAPO, a single-stage optimizer for the tight-budget regime. BudgetAPO incorporates (1) a noise-adaptive rule that sizes the evaluation slice to each task's noise, measured by a short probe; (2) a fixed slice that turns every accept/reject decision into a paired comparison; and (3) a reflective operator that rewrites reasoning strategy and output format jointly. Extensive results across seven benchmarks and five subject models demonstrate that BudgetAPO ranks first on every subject and beats every baseline under Holm-corrected paired tests, while returning the seed in 13% of runs at 250 calls against 86% for GEPA. On GPT-OSS-20B, GEPA needs 4.5 times as many calls to match BudgetAPO's 100-call score.

We study linear bandits with memory, where past actions induce endogenous nonstationarity through an arbitrary known, bounded matrix-valued memory map. To trade off exploration and exploitation while accounting for the memory dynamics, we develop RSM-LinUCB, a Bellman-centric algorithm that learns as in linear bandits and plans as in reinforcement learning. This design admits a novel regret decomposition which separates the memory-induced error from the cumulative reward estimation error along the learner's trajectory. We prove a high-probability regret bound of $\widetilde O\big(dRS(M+1)+σd\sqrt T\big)$, where $T$ is the learning horizon, $d$ is the parameter dimension, $M$ is the memory length, $R$ and $S$ bound the memory-map operator norm and reward-parameter norm, respectively, and $σ$ is the sub-Gaussian noise scale. Our results reveal that the multiplicative memory-horizon coupling in prior bounds is not intrinsic: memory only contributes an additive cost, up to logarithmic factors. We also prove a matching minimax lower bound, establishing near-optimality. We further extend the algorithm to generalized linear rewards, preserving this separation with near-optimal memory and leading statistical dependence. Our algorithms outperform the baselines in numerical experiments on synthetic instances and semi-synthetic KV- and semantic-cache tasks.

Graphs provide a natural representation of many complex systems, ranging from social platforms to ecosystems. However, the development of graph-based machine learning methods is often constrained by the limited availability of large and diverse graph datasets. In this paper, we introduce $\texttt{DCBA}$, a model-based approach to graph data augmentation that infers the configuration of a synthetic graph generator from an observed network. We instantiate the proposed framework using the $\texttt{ABCD}$ generator, which produces scale-free networks with community structure. Our model learns a joint representation of graphs and generator parametrisations using a multi-positive contrastive objective with soft negative weighting. The learned representation enables the prediction of an $\texttt{ABCD}$ configuration whose stochastic realisations preserve the macrostructural properties encoded by the generator. Experiments show that $\texttt{DCBA}$ recovers generator parameters more accurately and robustly than an algorithmic inverse-modelling baseline. Its downstream utility is further demonstrated in community detection, where inferred configurations used to fine-tune $\texttt{PRoCD}$ improve AMI on average by $161\%$ on synthetic and $273\%$ on real-world networks.

We investigate the scaling laws and compute-optimal training of physiological waveform foundation models (FMs). We train Aether, a family of over one hundred FMs ranging from 20M to 2.1B parameters, on up to 36.3M hours of physiological waveforms. We construct eight clinical prediction tasks from MIMIC-III and evaluate the FMs through linear probing. The 720M FM outperforms all existing baseline FMs across all eight tasks. A scaling law of model size, pretraining hours, and labeled patients predicts downstream ranking error, i.e. $1-\mathrm{AUROC}$, effectively with $0.5\%$ prediction MAE at held-out resource scales and $0.9\%$ MAE when extrapolating to 2.1B parameters. We present three findings: (1) Compute-optimal training scales both FM size and pretraining hours. Under the fitted law, a $10.0\times$ increase in compute FLOPs scales model size by $1.2\times$ and pretraining hours by $8.2\times$. (2) Larger FMs use waveform data more efficiently, and greater pretraining exposure increases the benefit of model scaling. Starting from 25M parameters and 4.8M pretraining hours, doubling FM size reduces the predicted hours needed for the same performance by $51.8\%$. (3) Pretraining and clinical supervision reinforce each other: more labeled patients increase the return to pretraining, while larger FMs and longer pretraining reduce labeling requirements. For the example of the 720M FM, extending pretraining from 120K to 36.3M hours reduces the predicted patient requirement by $61\%$ at a target ranking error. These findings provide a quantitative training recipe and a promising and durable scaling path for physiological waveform modeling and downstream clinical prediction.

The Lindblad master equation is the standard framework for describing the non-unitary evolution of open quantum systems, where environmental interactions induce dissipation and decoherence. When both the system Hamiltonian and the dissipation rates are partially unknown or explicitly time-dependent, traditional analytical inversion and system-identification techniques become intractable. Recent works have demonstrated that Transformer-based models can infer unknown dissipation rates from observable time series, yet these approaches typically rely on hand-crafted statistical features under idealized and highly restricted conditions. Here we advance the paradigm by introducing a raw time-series Transformer that directly ingests the full trajectories of Pauli expectation values $\langleσ_x(t)\rangle$, $\langleσ_y(t)\rangle$, and $\langleσ_z(t)\rangle$, thereby fully exploiting the self-attention mechanism for temporal modeling. The architecture is further extended to jointly learn unknown Hamiltonian parameters, handle multiple dissipation channels, and operate robustly under realistic measurement noise. Across all tested scenarios the model achieves consistently high reconstruction accuracy while eliminating manual feature engineering. This provides a scalable, robust, and versatile framework for quantum environment sensing in realistic open quantum systems.

A diffusion model starts from pure noise and removes it step by step. Somewhere along the way it stops being able to become "anything" and becomes committed to, say, a horse rather than a truck. We measure when this happens, and what a trained model gets wrong about it, on CIFAR-10. The most direct measurement is to freeze a half-finished image, restart the generation from that point many times, and count how often each class comes out. We call this probability the committor. Measured this way, the model settles coarse questions (vehicle or animal?) at roughly twice the noise level of fine ones (which animal?). A much cheaper measurement, the noise level at which a classifier's opinion about two classes splits into two distinct groups, gets the order of these decisions right (rank correlation 0.73-0.88) but not their exact timing. We then compare pretrained models with their

Emerging mission classes such as on-orbit servicing, satellite inspection, and active debris removal require trajectory design methods that are adaptable to a variety of mission scenarios. We present a diffusion-based trajectory generation approach for rendezvous and proximity operations (RPO) that enables flexible configuration of mission constraints. First, individual energy-based diffusion models are trained to satisfy distinct constraints such as approach cone and sensor line-of-sight from a set of optimized trajectories. Then, at inference time, the learned energy models can be composed with one another, or with an analytically defined energy field, to enforce specific constraint combinations. We validate this framework with the composition of a learned approach cone model and a learned sensor line-of-sight model, as well as a learned approach cone model and synthetic obstacle avoidance model, both of which yield constraint satisfaction rates that are within 1 percentage point of the single-constraint models or higher. These results indicate that our compositional diffusion framework can provide a modular approach to RPO trajectory design and enable reconfiguration for new constraint combinations without requiring model retraining.

Vision-language models (VLMs) are increasingly trained to generate structured outputs like points and bounding boxes that downstream interfaces, agents, and robots can act on, yet safety alignment of this output channel has not been systematically analyzed. We study visual grounding safety by repurposing three safety benchmarks spanning direct harm (VLSU), social bias (BBQ-V), and situational safety (Asimov-2.0) into 15,401 matched pairs of harmful requests that differ only in the requested output: a free-text answer (VQA) or a grounding (point or bounding box). Across five VLMs, models that refuse a harmful request posed as a question often comply when the same request asks for a grounding: averaged over models, grounding refusal trails VQA refusal by 31-59 percentage points, depending on the domain, and safety system prompts do not close this gap. We propose a fine-tuning approach that combines grounding-form refusals with capability grounding data and self-distilled benign data to counter over-refusal. For Qwen3-VL-8B and VisionReasoner-7B, it improves grounding refusal by 77-95 percentage points on VLSU and BBQ-V and by 64-85 points on the held-out Asimov-2.0 domain, while also improving VQA refusal, preserving grounding capability, and keeping over-refusal limited. Representation analysis shows that fine-tuning moves harmful requests toward each model's refusal direction, most strongly for grounding, while leaving benign requests near the harmless reference.

This paper presents GENAI4IDN Benchmark 3.0, the third iteration of an evaluation framework to assess Generative AI tools for creating Interactive Digital Narratives (IDNs). Moving beyond manual testing, this iteration introduces AI-assisted evaluation through a publicly accessible web application (https://genai4idn.com), enabling the community to run benchmarks on demand, add new models, and propose new tasks. The revised evaluation framework is "blinded" to avoid model-bias, and can handle complex media such as music, videos, and full IDNs that previously required human raters. A significant addition - responding to concerns raised during ICIDS 2025 - is the addition of fact-checking and bias detection with dedicated tasks and rubrics, validated by human raters with lived experience in the depicted contexts. Findings from a diverse range of models report on maturing creative capabilities while observing runaway thinking and overzealous safety filters as limitations. Fact-checking reliably caught subtle historical inaccuracies, anachronisms, and fabricated claims while the bias rater consistently exposed structural assumptions, tropes, and marginalized group erasures.

Partial observation and delay-coordinate reconstruction are rooted in deterministic dynamical-systems theory, whereas there are many systems with intrinsic stochasticity. We propose a conditional-moment interpretation of delay-coordinate reconstruction for stochastic systems, in which a delay vector is used to reconstruct conditional moments of a future distribution rather than a unique future sample path. Two complementary arguments motivate this viewpoint. First, the probability density of a stochastic differential equation obeys a deterministic Fokker-Planck equation and, under certain assumptions, is represented by an infinite deterministic hierarchy of moments. Hence, a finite-moment closure suggests a Takens-like finite-dimensional approximation. Second, a discussion based on the Koopman operator theory clarifies that the time evolution of an observable in the Mori-Zwanzig formalism yields a conditional expectation in stochastic systems. Then, the orthogonal "noise" term in the coefficient-space Mori-Zwanzig equation vanishes in the stochastic cases; this result is consistent with the moment-based argument. As an application of this stochastic delay-reconstruction viewpoint, we revisit convergent cross mapping (CCM) for diagnosing certain causal relationships. Although CCM based on the embedding theorem cannot generally be applied to stochastic systems, it is possible to examine certain types of causal relationships by using conditional moments. Using coupled logistic systems with additive and multiplicative coupling mechanisms, we discuss how causal relationships are embedded in stochastic systems.

Visual entailment (VE) asks whether an image supports, contradicts, or leaves undecided a textual hypothesis. Strong results come from fine-tuning large vision-language models on labelled data, while zero-shot and hybrid approaches remain far behind. A VE hypothesis often bundles several visual claims, yet existing zero-shot methods reason over it as a single unit. We propose Atomic Visual Entailment (AVE), which decomposes the hypothesis into atomic facts, produces candidate predictions from both the full hypothesis and its facts using frozen vision-language models, and predicts the final label with a lightweight classifier trained only on how those candidates behave. We find that decomposition helps only when the hypothesis context is preserved: judging facts in isolation is worse than not decomposing at all. Full-hypothesis and atomic prediction make complementary errors, and learning which to trust recovers far more of that complementarity than majority voting, reaching 0.803 test accuracy on SNLI-VE without fine-tuning any vision-language model. AVE also localises the visual evidence behind its prediction without region-level supervision. These results suggest that learning which candidate prediction to trust can close much of the gap to fine-tuned systems, offering a practical alternative where fine-tuning a vision-language model directly would need more labelled data or compute than is available.

This paper presents the workflow for building an automated ArcGIS Pro tool using ArcPy to extract the Land Surface Temperature (LST) from Landsat 7, 8 and 9 datasets. The tool eliminates the need for manual band selection and repetitive raster computations by automating the multi-step workflow of radiometric calibration, NDVI-based emissivity correction, and thermal conversion. In addition to supporting batch and single-scene processing, the tool has an optional Large Language Model (LLM) for statistical result interpretation and reporting. Depending on batch size, the tool reduced the processing time from around 11-58 minutes when done manually to around 4-11 minutes using the tool. We tested the tool with data from Ras Al Khaimah (RAK) in the UAE, and the LST obtained for Ras Al Khaimah ranged from approximately 25C to 50C, demonstrating an accurate LST mapping compatible with the weather conditions of RAK. In summary, our tool reduces human errors and improves processing accuracy and efficiency for thermal and environmental remote sensing applications, in addition to providing an interactive LLM-based interface for result interpretation.

Tool-using AI agents are increasingly deployed across enterprise software systems, yet widely used benchmarks primarily evaluate nominal task completion, conflating baseline planning competence with operational fault recovery. We introduce UndoBench, a benchmark spanning 36 base workflows and 36 fault scenarios across 8 enterprise domains, decoupling task competence from recovery capability via counterfactual paired trials under identical seeds alongside wire-level effect-history and environment-state oracles. On 12 held-out TEST workflows across two open-weight models, two frameworks, and three recovery paradigms (5,760 executions / 2,880 paired trials) in the frozen lost-acknowledgment study, nominal competence reached 83.54% while conditional recovery success rate (CRSR) fell to 46.72%, with naive retry producing duplicate external effects in 53.33% of trials. Extensions to commercial API models reproduced this competence-recovery separation. Evaluations across complementary execution boundaries show that recovery is phase-dependent: before mutation, methods perform similarly without duplicate effects among capable trials; during partial mutation, naive retry, per-call idempotency, and zero-privilege journaling collapse on the evaluated composite workflows; after commit but before acknowledgment, verification and server-side idempotency substantially improve safety. These findings demonstrate that evaluating nominal completion alone masks critical, phase-dependent recovery vulnerabilities in autonomous agents.

Finding which variables cause which others in multivariate time series, and at what lags, is central to science and policy, yet existing methods force a choice between flexible confounder adjustment, data-driven lag selection, and inference that controls the false discovery rate (FDR). ORACLE-VARX does all three in one pipeline. First, double/debiased machine learning (DML) removes nonlinear confounder effects from the outcomes and the lagged series. Second, adaptive causal lag estimation (ACLE) picks the lag order at each time step by sequential significance tests, tracking regime changes. Third, entry-wise $z$-tests with Benjamini--Hochberg correction select directed edges at a target FDR. We prove that in each rolling window, the debiased coefficients are asymptotically normal around a window-averaged target, so their $z$-tests are asymptotically valid. On a synthetic benchmark with time-varying structure and nonlinear confounding, ORACLE-VARX (LightGBM) tracks the true lag order best (RMSE $0.96$ vs $1.1$--$1.5$), has edge FDR $0.047$, close to PCMCI ($0.045$) and below VAR ($0.129$) and VAR-LiNGAM ($0.187$), and forecasts better than all three. On nine U.S. sector ETFs with macroeconomic confounders, it yields interpretable causal graphs whose lag order rises in high-volatility regimes.

Multi-agent debate (MAD) has emerged as an effective paradigm to improve the reasoning capabilities of large language models (LLMs) and is increasingly being extended to multimodal settings. However, existing multimodal MAD frameworks typically expose agents to the same fixed visual input, ignoring substantial variation in the visual scale needed across samples and agents. In addition, these frameworks frequently suffer from groupthink, a phenomenon where agents prematurely abandon correct deductions to conform with confident but hallucinated peer responses. To address these bottlenecks, we introduce DREAM (Dynamic Resolution Assignment For Multimodal Multi-Agent Debate), which operates via two core components: (1) Dynamic Resolution Assignment, a zero-shot probe round where agents test multiple resolutions, quantify uncertainty using Average Normalized Log-Likelihood (ANLL), and use an adaptive threshold to assign each agent to its empirically optimal resolution; (2) Uncertainty-Guided Rollback Aggregation counters groupthink by tracking each agent's uncertainty over rounds and restoring early low-uncertainty answers overridden by group pressure. On six multimodal datasets, DREAM improves the accuracy-token trade-off over multi-agent debate baselines by 1.5-3.2% accuracy without dataset-specific tuning.

Embodied, ego-centric intelligence fundamentally requires the ability to comprehend spatial audio within complex environments. While large audio-language models excel at mono-channel reasoning, they lack spatial awareness, discarding critical spatial cues that enable sound localization and that can improve the disentanglement of overlapping sound sources. To address this, we present SEA-LM, a Spatial Audio Understanding model. First, we introduce FOACODER, a layout-flexible spatial audio encoder trained on source localization and ego-centric voice activity detection objectives to encode First Order Ambisonics derived from variable-count, variable-position smart-glasses arrays via beamforming. We train a Multimodal Large Language Model (MLLM) to understand these spatial audio embeddings through a two-stage curriculum spanning six tasks, including sound localization and spatially selective transcription in settings with multiple speakers and overlapping sounds. To prevent the transcription outputs from dominating the next token prediction loss and overwhelming the direction predictions, we introduce a Spatio-temporal Weighted Cross-Entropy Loss. On our evaluation set, SEA-LM achieves lower azimuth and elevation MAE, higher temporal IoU, lower external-source hallucination and missing-source rates, and lower WER on most transcription tasks than compared baselines, while remaining robust across 1,211 smart-glasses array configurations with 4 to 9 microphones.

We present Kandinsky 6.0 Video, a family of foundation diffusion models for synchronized text-to-audio-video generation, comprising Kandinsky 6.0 Video Lite (3B parameters) and Kandinsky 6.0 Video Pro (29B parameters). Both models generate 5-second video clips with synchronized 44 kHz audio, including lip-sync, in text-to-audio-video (T2AV) and image-to-audio-video (I2AV) modes; a built-in super-resolution model raises the output resolution to Full-HD (1920$\times$1080). Building on the video generation capabilities of Kandinsky 5.0, Kandinsky 6.0 Video employs a dual-stream CrossDiT architecture that connects a pretrained video stream and a newly trained audio stream through bidirectional cross-attention for temporal and semantic alignment. Our continuous pretraining strategy first trains the audio stream from scratch on large-scale audio corpora and then trains both streams jointly on paired audio-video data while preserving unimodal fidelity; pretraining is followed by supervised fine-tuning, reinforcement-learning-based post-training, and distillation. In side-by-side human evaluation, Kandinsky 6.0 Video Pro clearly outperforms its predecessor, Kandinsky 5.0 Video Pro, and remains competitive with leading audio-video generation models, particularly in speech quality. To accelerate open research and deployment in multimedia generation, we release the code, model checkpoints, and diffusers integration under the MIT license.

Existing attacks on unlearned diffusion models assume that the erased concepts are known in advance and focus on recovering them. In practice, however, model providers may not disclose which concepts have been removed, and even with access to the original base model, an adversary may still lack a clear target to attack. In this paper, we aim to answer the following critical but overlooked questions: which concepts have been erased from the model, and how many have been erased in total? To this end, we present Tracer, a framework that rapidly and accurately identifies erased concepts and estimates their number. Tracer efficiently identifies erased concepts without generating and classifying images. By combining lightweight spectral analysis of weight footprints, it enables efficient search over large candidate vocabularies. To distinguish multiple erased concepts, we introduce a footprint coverage objective that guides sequential discovery. Tracer estimates the number of erased concepts by detecting a sharp decline in candidate confidence as the selected concepts account for the erasure footprint, without requiring labeled examples for calibration. The framework requires only lightweight linear algebra and limited forward probes, with no prior knowledge of the unlearning algorithm. Experiments across text-to-image and text-to-video backbones and diverse unlearning methods demonstrate that Tracer identifies erased concepts and estimates their number in seconds, achieving 150 to 137,000 times and 133 to 20,000 times speedups over MIA and brute-force search on image and video models, respectively, with substantially higher identification accuracy.

High-dimensional bioinformatics data, characterized by a large number of features relative to the number of samples, pose major challenges such as the ``curse of dimensionality,'' leading to overfitting, high computational cost, and poor generalization. Traditional feature selection methods often suffer from limited scalability and adaptability in such domains. We propose an LLM-in-the-loop reinforcement learning (RL) framework for bioinformatics feature selection, where the RL agent formulates feature selection as a sequential decision-making task, while the large language model (LLM) enhances the process in two ways: (1) guiding exploration through domain-informed advice, and (2) providing hybrid rewards that integrate data-driven performance with knowledge-driven evaluation. The LLM also produces explanations to improve interpretability for human experts without altering the RL policy update. Experiments on diverse bioinformatics datasets show that the LLM-in-the-loop framework outperforms baselines, achieves stable performance across downstream models, and converges faster than pure RL.

Temporal dependence can separate the Gaussian approximation of stochastic gradient descent from its stationary moments. For unmodified least-squares SGD, we construct a design with standard Gaussian marginals whose stationary error has every positive moment infinite. Independent observations with the same marginals instead give finite stationary variance. Both regimes retain a Gaussian small-step limit. Our general theory establishes pathwise contraction from a finite second design moment, then uses score cancellation and localization to obtain stationary Gaussian and Ornstein--Uhlenbeck limits. Independent Gaussian regression errors yield an exact conditional Gaussian law and total-variation convergence under the same design integrability. Stronger design conditions identify a positive first-order total-variation constant and a deterministic covariance correction with $o(a)$ error. A scalar coverage expansion translates this correction into its inference consequence. Experiments examine distributional error, coverage, and calibration with dependent scores. Together, these results establish precise probability-law approximation beyond moment-based stationary analysis.

An organization often holds too little labeled data to train a model that generalizes, and the records that would supply the rest sit with organizations that cannot release them. Cross-silo federated learning offers a way through, since participants exchange model parameters rather than records, but it ordinarily settles two aspects of the arrangement in advance, the participating sites and the classes the model can predict, and deployment can breach both. A new site joins after training, once the established sites have finished their engagement and gone offline, and its records arrive unlabeled, mixing conditions the model already recognizes with conditions no participant has observed. We present an autonomous three-stage procedure that expands the model entirely at the joining site: reconstruction experts screen for novelty, clustering separates the flagged records into candidate conditions, and class means describe the old classes, all inside one shared representation. Those classes were learned from records that never leave their owners, so the usual defenses against forgetting are unavailable, and the procedure supplies the evidence they would have carried from either of two dissimilar sources, prototypes held by the federation or records held by the joining site. On a real industrial condition-monitoring dataset, run end to end with no label consulted, either source holds old-class accuracy at 0.868 or above with forgetting at most 0.063, and the two differ by 0.021, so a configuration can be chosen by the disclosure it permits rather than the accuracy it delivers. Both keep old- and new-class accuracy in balance where every alternative we measure gives up one for the other, and both retain more of the old classes than distillation- and regularization-based baselines. The balance still holds with only 6 labeled records per arriving condition and 3 retained per old class.

Language models process and generate text sequentially in token units, and the tokenizer determines how much text each inference step covers. Under standard tokenization, a short English phrase such as "On the table." is usually produced as four separate predictions for the preposition (On), article (the), noun (table), and punctuation (.), where each consumes a sequence position and adds inference cost. We introduce CoBPE, a compositional tokenization approach that represents such phrases as a lexical base token (table) attached with a small set of reusable surface modifiers, composed in embedding space at input and predicted jointly at output. In controlled pretraining from scratch at 780M and 1.3B scales, CoBPE shortens sequences by 30% and improves average downstream performance by 1.2 points relative to standard BPE under matched training compute. Our results suggest that part of what is now expressed through token sequences can instead be modeled through structured representations, opening a broad design space for more token-efficient and capable language models.

Local Gaussian models of constant-step learning predict output variability and expected losses, but weak convergence alone does not justify these moment predictions. We establish moment-accurate Gaussian mixtures by matching stationary energy with local Ornstein--Uhlenbeck limits, ruling out quadratic tail mass invisible to weak convergence. For step size $a$, the second-order Wasserstein error is $o(\sqrt a)$, uniformly over invariant laws, using each law's actual root weights. The assumptions combine confinement, descent, finitely many hyperbolic equilibria and root continuity with finite-variance innovations. The result yields observable covariances, expected objective gaps and first-order mean shifts, while allowing singular covariances, compatible saddles and weights without a limit. For additive noise given by a fixed invertible transform of independent standardized Student $t_3$ coordinates, symmetry gives an order-sharp $\sqrt a$ smooth-test bound. Numerical transport calculations demonstrate the value of root-specific covariances; controlled SGD studies assess observable predictions across step sizes, batch sizes and model geometries.

As pretraining increasingly repeats data, every run faces three questions: how many epochs to take, how that number should change with model size, and whether anything besides the epoch count matters. We answer them by pricing a repeated token against two references: one epoch on the same data, which gives its value, and fresh data at equal compute, which gives its cost. Against fresh data, the cost of repetition follows a single variable, the number of extra epochs divided by the unique tokens per parameter. Against the same data, a second epoch is worth nearly as much as a fresh one, and repeated tokens fall to half the value of fresh ones after a critical epoch count that grows with the training budget per parameter but hardly with model size. With unique data fixed, the predicted compute-optimal run grows model size and epochs together until loss stops improving, near the critical epoch count. The same variable accounts for the direction of size trends that appear to conflict: larger models tolerate fewer epochs when the corpus is fixed, from about 15 at 127M to 4 at 2B parameters, but not when unique data grow with the model. Counts alone do not determine loss: at identical counts, replaying shards consecutively raises loss by up to 0.46~bits per byte, concentrating repeats on fewer samples also raises it, lower-entropy sources degrade faster with repetition, and re-tokenizing repeats helps only under heavy repetition. These results offer an empirical guide to pretraining when unique data, rather than compute, are the binding constraint.

Cold-start drug-drug interaction (DDI) prediction tests whether models can identify clinically significant interactions for drugs without training-time interaction history. Existing benchmarks mostly report aggregate edge-prediction scores, leaving a key evaluation question unanswered: when models receive molecular, textual, or knowledge-graph (KG) evidence, do they actually use the evidence that pharmacologically supports the interaction? We introduce ColdDDI, a reconstructible diagnostic benchmark built from DrugBank 5.1.13, with 1,900 approved small-molecule drugs and 565,731 positive DDI pairs. ColdDDI evaluates pairs with zero, one, or two unseen drugs. It also annotates each interaction by whether it changes drug exposure or drug effect, and by whether the biomedical knowledge graph contains shared enzymes, transporters, or targets that can plausibly mediate the interaction. These annotations separate evidence availability from predictive dependence. We evaluate eight conventional DDI methods and 13 LLMs; for open-weight LLMs, we test five prompt patterns and use masking, drug replacement, and channel-sensitivity metrics to probe knowledge utilization. ColdDDI exposes that, in the hardest split where both drugs are unseen, the main performance divide is mediator availability. A fine-tuned 1B LLM recovers 89-93% of interactions with a shared enzyme, transporter, or target, but only 40-62% without such a mediator. More importantly, KG-provided evidence is not always used; several KG-augmented baselines change little when the shared mediator is masked or disrupted, whereas fine-tuned LLMs respond strongly to this intervention. Thus, ColdDDI evaluates knowledge utilization rather than knowledge access alone, showing where cold-start DDI models rely on mechanistic evidence and where they fail despite receiving it. Code is available at https://github.com/0217ljh/ColdDDI-NeurIPS2026.

As Large Language Models (LLMs) gain tool use capabilities such as web search, they can retrieve and cross-reference public information, creating privacy risks beyond memorization. One manifestation is re-identification: linking an anonymized interview transcript to a named individual. Yet without ground-truth identities, the coverage of such attacks and the protection offered by a defense cannot be reliably measured. We introduce AgentDOXX, an evaluation suite of 822 synthetic interview transcripts grounded in public information about real individuals with known identities. We evaluate fifteen configurations of open-weight and proprietary models, isolating the effect of web search, and analyze their search trajectories to distinguish retrieval-driven from parametric identifications. Ground-truth identities reveal that re-identification risk is distributed across an agent's execution: retrieval and parametric recall both contribute, with open-weight models identifying 15-28% of transcripts without search; identification succeeds in over 88% of cases once the target appears in a retrieved result; entity masking leaves at least one attacker successful on 85.3% of a stratified sample; and privacy instructions suppress naming but not retrieval, with configurations scoring 0% accuracy yet retrieving the subject in up to 62% of transcripts. We further show that observed attack trajectories can provide supervision for localizing identifying spans, offering a path toward attack-informed anonymization.

Extra inference compute is usually spent on sampling more reasoning chains. We study where inside an existing chain an additional continuation should begin. We define expansion utility, the change in correctness from restarting a chain at a stored step, and measure it at every eligible step for nine models on six benchmarks (41 model and benchmark cells). Restart position matters: steps selected on one set of continuations beat uniform placement when scored on disjoint ones, in held-out audits on 5, 16, and 38 cells (+4.25 points [+2.51, +6.63] in a fresh five-cell audit). A fixed rule that restarts from the last eligible steps, always-last, is a strong baseline: our learned router beats uniform placement but shows no detected gain over it, and on DeepSeek-R1-Distill-Qwen-14B/MATH-500 always-last exceeds the exact self-consistency frontier at matched aggregate generated output by +0.052 [+0.008, +0.098], using 0.774x the aggregate generated output of four-sample self-consistency. Cross-fitted oracle selection still finds held-out headroom beyond declared positional classes, a target for future selectors. Finally, breaking step-label ties by earliest index flips the sign of a pointwise selector's gain over uniform placement in every seed of a five-seed diagnostic with four rollouts per step; randomized ties remove the bias.

A linear probe can show that a label is recoverable from a model's hidden states, but not whether that goes beyond what the input already reveals, or whether the model uses it. We audit Toto, an observability forecasting foundation model, on the Benchmark of Observability Metrics (BOOM) across five series-disjoint resplits, comparing linear probes on its frozen residual stream with models that read the raw input window and with Toto's architecture stripped of its trained configuration. Short-vs-medium cadence and metric type are more linearly recoverable from Toto's residuals than from the strongest raw-window model in every resplit (macro-F1 0.766 vs. 0.633 and 0.545 vs. 0.498). Domain is nearly tied, and series cardinality is recovered far better from the raw window. MOMENT-base shows related cadence, metric-type, and domain readouts. Recoverability is not use: exchanging Toto's residuals with those of high-burst donors moves a future-burstiness readout as intended but does not make forecasts consistently burstier than a randomized donor. A BOOM-trained coordination probe has negative zero-shot R^2 on the tested external benchmarks. We report each label against its strongest baseline.

Much of expressive text-to-speech research rests on an untested assumption that written text carries enough information to select an appropriate prosodic style for its delivery. Text-predicted style models improve listener preference, and expressive-appropriateness evaluation presupposes that context constrains style, yet neither measures the assumption itself. This paper tests it as a falsifiable hypothesis against style labels derived from acoustics alone. For each of six speakers in a 1,200-hour conversational corpus, utterances are clustered in the spaces of five speech models, including a prosody-only control, and the cluster of held-out utterances is predicted from twelve text embedding models. Three controls are applied: utterance length is erased from the speech embeddings; accuracy is scored against the majority-class floor of unbalanced clusters rather than uniform chance; and a bag-of-words baseline measures word identity alone. Text predicts the cluster above that floor for all six speakers (+0.111 top-3 accuracy), but bag-of-words achieves three quarters of this. Sentence embeddings add only +0.026, largest for encoders not trained for sentence semantics and reversed by tree-based probes for all others. Acoustic clusters are not compact in text embedding space in any of 360 configurations. The prosody-only space weakens the association for five speakers, but not for the speaker showing it most strongly. Text thus informs these delivery clusters mainly through word choice, whether as a cue to prosody or as a marker of topic and recording situation, and reference-free style selection cannot assume more.

Improved name retrieval may have little effect on company clusters when the pair classifier remains unchanged. We examine this dependency by adapting multilingual E5 encoders under fixed candidate budgets and downstream decision rules. Random-negative and hard-negative training use identical positive schedules. Checkpoints are selected before collecting a new GLEIF sample of 3,633 names, 2,880 source identities and 882 silver-positive pairs. At 72,660 candidate edges, adaptation with a multi-view selector increases direct pair recall from 53.74% to 76.98%. The primary matcher adds only seven correct and two incorrect co-cluster pairs: cluster recall rises from 32.54% to 33.33%, while precision falls from 95.99% to 95.45%. Of 208 newly retrieved silver-positive pairs, 202 fall below its decision threshold. Random-negative and hard-negative training produce identical final partitions. An AI-assisted, single-reviewer audit of 137 pairs supports the observed pattern, although its predominantly LEI-derived evidence does not establish independent gold labels. The results locate the immediate loss of retrieval gains at the existing confirmation stage and show why encoder evaluation must also measure final cluster quality.

Predicting whether an LLM agent will fail has emerged as a promising direction for supporting intervention during execution. Recent approaches report strong predictive performance, often with AUROC values between 0.85 and 0.94. However, predictors are typically trained by pooling runs from many tasks. We hypothesize that part of this performance comes from recognizing that some tasks are harder than others, rather than detecting whether a particular run is heading toward failure. This distinction matters because task-level difficulty supports decisions about where to allocate computation, while run-level prediction is needed to decide whether to intervene in an ongoing trajectory. We study benchmarks with repeated attempts of the same task by the same agent and separate cross-unit comparisons from comparisons between successful and failed runs of the same model-task unit. Across the evaluated corpora, more than 99.93% of the positive-negative pairs underlying pooled AUROC are cross-unit. Accordingly, predictors that never observe the current run can achieve high pooled performance, including a difficulty oracle with AUROC up to 0.945, while remaining at chance within task. Early run-level discrimination is consistently weak across trajectory predictors, released monitors, and hidden-state probes, although it improves later in execution and is stronger for weaker agents. Under fixed token budgets, task-level allocation outperforms abort-only strategies, while early stopping becomes beneficial only when within-task AUROC reaches about 0.84-0.93, far above the 0.50-0.55 range observed for early monitors. These results show that failure prediction should be evaluated not only by outcome accuracy, but by whether the captured signal supports the intended deployment decision.

In safety-critical domains such as autonomous driving, systems must be evaluated across a large number of environment conditions, often represented as composite scenarios built from primitive scenarios. Existing statistical model checking (SMC) approaches analyze each composite scenario independently, requiring many expensive simulations and resulting in substantial redundant computation when scenarios share common structure. This work introduces a scenario-based compositional SMC framework for safety and co-safety specifications, enabling efficient analysis of composite scenarios. Our approach decomposes scenarios into primitives and specifications into sub-specifications, verifies each primitive independently, and composes the resulting statistical estimates using importance sampling and kernel density estimation. Our empirical evaluation shows that the proposed framework can accurately answer verification queries for previously unseen composite scenarios while reducing simulation cost through parallelization and trace reuse.

Existing thermodynamically consistent neural operators impose the GENERIC degeneracy conditions by projecting the reversible operator onto the complement of the entropy gradient. This makes the operator state-dependent and forfeits the Jacobi identity, so the result is metriplectic-degenerate rather than metriplectic. We instead obtain degeneracy the way GENERIC does. For nonlinear transport, the reversible operator $L$ is the compatible Lie-Poisson pencil $αD+λ(uD+Du)$; otherwise it is a constant, trivially Poisson Fourier multiplier. On an augmented state $(u,s)$ with a latent entropy density, $S=\int s$ is a Casimir of $L$, so $L\,δS/δz=0$ holds identically without projection. The energy combines a fixed mechanical quadratic, a learned gauge-free potential, and a convex internal energy. The friction operator $M=AA^\top$ satisfies $M\,δE/δz=0$ pointwise, and its Onsager parity structure permits diffusion and damping while provably excluding transport. For any parameters, skewness, positivity, both degeneracies, and the Jacobi identity (on the resolved band for the Lie-Poisson term) hold to machine precision. Heat conduction and damped waves admit exact closed-form friction operators, the second law bounds physical energy under a checkable curvature condition, and a discrete-gradient integrator yields exact discrete first and second laws. On four PDEs in 1D and 2D with three backbones (FNO, Transolver, CNO), the model wins 61 of 72 seed-level comparisons against same-backbone unconstrained baselines, learns the exact transport and wave symbols, matches the true dissipation rate within 13% on heat and Burgers, and dissipates nothing on advection. A constant-$L$ ablation isolates the cost of exact Jacobi as the loss of Burgers, while a learned-entropy ablation injects energy on every reversible-irreversible problem.

We introduce Factoriax, a GPU-accelerated factory-building simulator written in JAX. In Factoriax, an agent must collect resources, build machines using those resources, and then automate the collection and crafting process by arranging machines on the map to build production pipelines. This paper discusses the structure of the Factoriax simulator and an initial benchmark called Easy Rocket in which an agent is tasked with building a resource-intensive machine called the Rocket in a limited number of game ticks to escape the planet. We also publish results from a number of PPO-based training runs on Easy Rocket. Factoriax is built to be fast. A 1-billion-step PPO training run, equivalent to 500,000 episodes, runs on Easy Rocket in about 8 minutes on a single NVIDIA A100. A standard laptop GPU can complete the same run in about 84 minutes. Our trained PPO agent learns to gather resources, craft machines from those resources, and place them on the map through a curriculum reward directly tied to a manually designed set of achievements. After training, the agent does not place machines in a functional spatial configuration, failing to fully complete the benchmark, and leaving the challenge open for future attempts.

Human health is a continuous, dynamic trajectory shaped by the cumulative interplay of biological processes, clinical events, behaviors and environmental exposures across the life course. Unifying the full breadth of lifelong health information, including longitudinal records, genetic variation, molecular profiles and environmental histories, is essential for whole-person modeling and remains a major challenge. We introduce LifeLong Digital Twin, a unified, event-driven modeling paradigm that organizes Life Events into daily Health States and accumulates them into Lifelong Health Context. An accompanying Agent Harness incorporates multimodal evidence beyond the language model's textual context. We evaluate four language models across 25 disease endpoints on three tasks: Disease Trajectory Forecasting, Disease Risk Ranking and Multi-horizon Disease Prediction. The approach yields marked gains over the reference condition: model-averaged F1 increases by 22.0% for disease identification in trajectory forecasting and 18.3% for five-year disease outcomes; thyroid-disease F1 reaches 0.669. The framework provides a foundation for whole-person digital twins and research on personalized lifelong disease prevention.

Text embedding models trained with contrastive learning learn to follow task instructions from instruction-paired data, while instruction-tuned LLMs already know how to follow them. We show that this instruction-following ability can carry over from an LLM to a Transformer-based embedder without any training. We propose Attention Relay, which passes the attention weights an LLM produces to the embedder's own attention. Across six instruction-tuned LLMs from the Qwen3, Llama 3.1 and OLMo 3 families and ten widely used embedding models that differ in tokenizer, size and pooling type, Attention Relay makes nearly every combination instruction-aware. Experiments that break the method down into its parts show that the LLM's attention weights track the instruction in its later layers and come largely from instruction tuning. They also show that relaying these weights selects which content in the text matters: it makes the aspect of the text that the instruction asks about dominant in the embedding, or restores that aspect where averaging had diluted it.

Small language model (SLM) agents need safety controls that track consequences across tool calls with little monitoring overhead. A private read, for example, becomes a leak when a later action sends that data outside the system. We introduce CARB (Conformal Agent Risk Budget), which calibrates when to stop an agent using a ledger of harm incurred before stopping. Under exchangeable episodes, standard conformal risk control bounds this declared loss in expectation over calibration and a future episode. G-CARB selects scorer evidence along observable dependencies from private sources to outgoing actions. The ledger still covers the entire executed history, and computing the gate score requires no additional language-model inference. On AgentDojo replay with two 14B backbones, G-CARB roughly halves scorer-input records at intermediate risk budgets while improving autonomous task completion relative to full-prefix scoring; random context of the same size achieves similar gains. Controlled examples show how retaining the relevant dependency can further avoid stopping benign work.

Industrial sequential recommender systems operate over massive item catalogs (e.g., 10^5--10^7 items). Multi-label recommendation models are trained with Binary Cross-Entropy (BCE) loss over the full vocabulary, but standard BCE materializes a dense [B, N, V] logits tensor in High Bandwidth Memory (HBM), incurring prohibitive $O(BNV)$ memory and fatal Out-Of-Memory (OOM) errors. While chunked loss optimizations exist for Softmax Cross-Entropy in LLMs, large-scale multi-label BCE optimization remains unexplored across deep learning ecosystems. We propose CutBCE, an exact, hardware-accelerated BCE loss and gradient operator implemented in JAX and Pallas for large-vocabulary workloads. CutBCE introduces (1) an exact fused reformulation evaluating dense background loss and sparse target corrections; (2) a custom Vector-Jacobian Product (VJP) with a dedicated Pallas TPU backward kernel computing logit tiles on-chip in both passes so logits and their gradients never reside in HBM; (3) dynamic VMEM budgeting and sharding-aware collective hoisting for distributed meshes; and (4) count-based zero-overhead training metrics. On single-chip TPU v5e/v6e mini-benchmarks, CutBCE eliminates OOM errors with up to 91.9% speedup. On 8-chip TPU slice training for multi-label SASRec with 876k items (Yambda-50M), CutBCE reduces peak HBM by 65.7% (>14 GiB saved per chip) and increases training speed by 225.9% with comparable accuracy. CutBCE is open-sourced at https://github.com/AI-Hypercomputer/RecML/blob/main/recml/core/ops/binary_cross_entropy_ops.py.

The Dobrushin coefficient of each attention-rollout factor satisfies $κ(\frac12(I+A))\le\frac12(1+κ(A))$, and multiplying these inequalities over layers bounds the coefficient of the whole rollout. We characterise exactly when the layerwise bound is tight: equality holds if and only if some token pair attaining $κ(A)$ is mutually self-dominant - each of the two attends to itself at least as strongly as the other attends to it. The condition is far from automatic: uniformly random stochastic matrices satisfy it only 24-30% of the time. When tested on the head-averaged attention of each individual input and restricted to content tokens - image patches, words or tabular features, excluding cls, register and separator tokens - the condition holds for essentially every input at every layer of DINOv2 (three model sizes), RoBERTa and DistilBERT. In the supervised models DeiT-B and ViT-B/16 it holds for 91% and 64% of input-layer pairs respectively, with all failures occurring late in depth. In FT-Transformer trained on two standard tabular benchmarks it holds for only 11-44% of input-layer pairs. The special tokens account for almost all failures in DINOv2 and the language models: when they are included, the condition holds for only 82-97% of input-layer pairs.

The component that dominates a latent world model's prediction error need not be the one whose repair most improves action selection. We show this by comparing action sequences from identical physical starts and separating endpoint error into a candidate-pool center and action-relative responses. Across four model families and four tasks, a confirmation pool of 256 new starts per task and 300 shared candidates per start shows that center error dominates MSE in 14/16 model-task cells. Yet in six of these cells, an oracle that corrects only the action-relative responses yields better physical rank correlation and top-30 elite quality than one that corrects only the center, while leaving more latent MSE (family-wise corrected intervals). The preference differs across the evaluated settings: a separate LeWorldModel (LeWM) study that executes oracle-selected actions favors center repair on PushT and on Reacher with a render-matched goal. Matched-candidate tests localize ordering loss: for LeWM, encoding realized endpoints raises physical Spearman from 0.464 to 0.975 on that Reacher setting and from 0.193 to 0.631 on PushT (64 starts per task), while Cube's encoded-goal cost remains uninformative. A 72-run objective study improves selected response diagnostics, while incremental closed-loop planning gains remain unconfirmed. These results separate error magnitude from the decision effects of oracle correction and motivate evaluating representation, prediction, and planning as separate stages.

We formulate joint estimation of common-slope decay rates and amplitudes from room impulse responses (RIRs) as parameterized nonnegative matrix factorization with the Itakura--Saito divergence as the loss function (IS-NMF). Estimation at each short-time Fourier transform frequency bin produces detailed reverberation time (RT) curves directly from RIR powers with no backward integration needed. Standard space-alternating generalized expectation-maximization (SAGE) algorithm yields closed-form amplitude updates and a convex subproblem for each decay rate update. To accelerate estimation, we introduce contribution-weighted SAGE, which emphasizes observations where each component contributes strongly to the modeled power. Experiments with synthetic data show accurate recovery of well-separated decays and faster loss reduction than standard SAGE. Application to measured coupled-room RIRs yields frequency-dependent RT curves and reveals complementary space-time contributions of the shared decay components.

Digital libraries, repositories, and AI-mediated knowledge services increasingly rely on generative systems to produce summaries, descriptions, and other multi-claim knowledge objects. Yet generation can scale far more easily than verification capacity: a reviewer may need to decide whether an object is suitable for publication or downstream use after checking only a small fraction of its claims. This creates a fundamental gap between claim-level verification and confidence in the object as a whole. The central question is therefore what a successful partial check implies about the reliability of the complete artifact when its size grows but the verification budget does not. This paper contributes a Bayesian model of bounded verification centered on the Predictive Value of Pass (PVP). The model predicts that evidentiary value decreases as artifacts grow under fixed verification capacity, improves with larger verification budgets, and is especially fragile when errors are sparse. Controlled experiments on FEVEROUS and FEVER support these predictions and show that adding supported claims around a fixed number of false or unsupported claims can make passing more likely while making a pass less informative. The model further yields the minimum verification budget required to maintain a target PVP, providing a practical component for AI-assisted quality-assurance workflows in which generated knowledge objects must be checked before publication or downstream use.

Real-world deployment of active learning typically forces practitioners to choose an acquisition strategy before any data is labeled. This is a daunting task: strategy performance varies widely across settings (e.g. datasets, surrogate models) and cannot be assessed without deployment. Existing strategy selection methods explore one strategy from a portfolio at each round and identify the optimal one using bandit feedback or model retraining. Many acquisition rounds are therefore spent exploring strategies rather than collecting the most informative data. Such overhead is a major barrier to AL-driven design of high-throughput experiments, such as genetic perturbation screens and directed evolution, where AL runs consist of only a few rounds with large batch sizes. This regime permits a natural alternative: acquiring data using multiple AL strategies within a single batch. We refer to this as soft strategy selection and introduce FractAL, a method specifically designed for this task. FractAL infers a per-strategy reward using influence-function-based data attribution, which requires no additional retraining, and then computes budget shares for each strategy in the portfolio using online mirror descent. We benchmark FractAL across 7 setups spanning classification, regression, and genetic perturbation effect prediction. The results highlight that strategy selection is a hard problem: every existing method performs worse than random sampling on at least one setup. FractAL, however, matches or outperforms every baseline, including random sampling, on all 7 setups. Its allocations concentrate budget on the strongest strategies in the portfolio while pruning the weakest. FractAL is therefore a reliable choice for real-world deployments, where the optimal strategy is unknown in advance, an important step towards making AL practical for high-throughput experiments and modern scientific discovery.

Mechanistic interpretability has emerged as the primary means to understand safety behavior of LLMs. However, existing tools primarily focus on the activating neurons or features of a model. The role of the remaining large set of inactive components is invisible to such methods. This work demonstrates that the inactive set contains safety-critical features that are causally relevant for refusal of harmful prompts. Suppressing such features could turn refusals into compliance, while passing undetected by prevalent interpretability tools. We introduce the Counterfactual Activation Potential (CAP), a metric that quantifies a suppressed feature's latent activation tendency as the product of its encoder alignment (how strongly the input drives it), suppression strength (how strongly active features inhibit it), and safety criticality (how much refusal depends on it). To find suppressed safety features at scale, we propose CAP-guided Safety Feature Discovery (CSFD), a two-stage filtering algorithm that identifies candidate safety features from hundreds of thousands of transcoder features without exhaustive ablation. A significant fraction of trials turn compliant with harmful prompts when a candidate feature is ablated. Under natural jailbreaks, the suppression acting on the highest-CAP features rises 2-4x, and their activation correspondingly falls by up to 80%. Amplifying a feature's suppressors pushes its activation down and raises harmful compliance with prompts related to the suppressed feature, with no such effect for random features. Our experiments span five Gemma, Qwen, and Llama models across various parameter sizes. Our findings indicate that jailbreaks could operate in part by suppressing safety-critical features rather than solely activating harmful ones, and that suppressed features are a necessary complement to activation-focused interpretability of safety behavior.

Natural language to SQL (NL-to-SQL) benchmarks are foundational to progress in data analysis research, yet recent work has shown that widely-used benchmarks contain significant annotation errors. These errors silently corrupt evaluation metrics, penalize correct model output, and distort the field's understanding of state-of-the-art performance. We present SALUS, a system that automatically detects annotation errors in NL-to-SQL benchmarks. SALUS frames benchmark auditing as a weakly supervised error detection: SQL generated by multiple LLM agents drive a suite of complementary weak-labeling functions. By passing this noisy vote matrix through a generative label model, we extract high-confidence training samples without requiring human ground truth. These samples train a decision plane that maps gold SQL query features to per-agent trustworthiness, allowing SALUS to intelligently fuse reliability estimates with raw verdicts for rigorous benchmark error detection. We evaluate on BIRD-Clean-xs, a benchmark of 298 BIRD development tasks with manually verified correctness labels. SALUS achieves F1 = 0.9194, significantly outperforming the state-of-the-art baselines. Applying SALUS to the full development sets, we estimate annotation error rates of approximately 37% on BIRD and 27% on Spider.

We introduce Loop Flow Transformers (LiFT), a family of looped generative models that scales computation by repeatedly applying a shared Diffusion Transformer (DiT) core, with only light changes to the standard architecture. Rather than asking every recurrent step for the final prediction, LiFT trains each step with a single regression target: a point on a straight path from the model's initial estimate to the flow-matching target. Because we index these targets by a continuous depth coordinate, a trained model can loop far beyond its training depth with no retraining, early exits, or other modifications. In our experiments, these longer rollouts improve generation, so inference computation can grow without adding parameters. On ImageNet at 256x256, LiFT-L/2 achieves an FID 3.34 points lower than our dense DiT-XL/2 baseline while using approximately 60% fewer parameters, 32% fewer training FLOPs, and 52% fewer inference FLOPs.

Human--LLM interaction datasets shape our understanding of AI use and provide a foundation for downstream research, including training and evaluation of user models. In recent years, a growing number of datasets have sought to capture a representative picture of human--LLM interactions. But how different are the pictures these datasets provide, and what do those differences mean for research built on them? We study these questions across seven conversation datasets, spanning in-the-wild chat logs and human preference data. We begin by revisiting the dataset classification experiment of Torralba & Efros and find that neural network classifiers identify the source of a conversation from user messages alone well above chance, indicating distinctive dataset signatures. This separability persists after matching datasets on the dimensions of human-designed taxonomies, implying subtle differences that these taxonomies do not capture. We then examine the implications for user modeling: how dataset signatures propagate to the outputs of user models trained on these datasets; how dataset choice influences evaluations of user model quality and subsequent evaluations of LLM assistants paired with these user models; and how dataset classifiers can guide data selection for training user models. While each dataset is meant to capture a slice of 'real-world' interactions, our findings reveal the extent to which these slices diverge, and the consequences of those differences for research built on these foundations.

Harness search keeps a change to the prompts, reasoning switches, token budgets or parsers around a frozen model if the change raises a score. Such a gain can come from answers the parser could not read before, a weak comparison, or a defect in the evaluation. We preregistered a study of where these gains come from, with three small models, three benchmarks, replication and test partitions, a GEPA search arm and six evaluation defects injected one at a time, and we report all 47 primary endpoints. Turning thinking off raised accuracy over a capped thinking setting in 5 of 9 model-benchmark cells, and in each the gain came mostly from questions where the capped setting gave no readable answer. The thinking-off setting was not meaningfully worse than a rescue configuration or four GEPA-selected harnesses in 11 of 13 comparisons, and lost to the rescue on GSM8K for two models. GEPA repaired its broken starting points, but none of its selected harnesses was more accurate than the thinking-off setting. A thinking budget in the serving engine, which also allows a longer answer, lowered truncation and raised the parse rate in 6 of 9 cells. In 6 of 15 evaluable defect-model pairs, replication through the same pipeline reproduced the defect's distortion instead of revealing it. On the LongevityBench multiple-choice tasks, only the longevity-tuned model beat the strongest constant-label baseline.

AI agents that send emails, edit files, and make purchases must decide when to act on their own and when to check with the user first. This decision is usually evaluated by showing a model a proposed action, asking whether it should proceed, and scoring agreement with human labels. We introduce DelegationBench to test whether such scores can be trusted. It has 156 scenarios with four possible responses (act, ask for permission, ask for missing information, refuse), and most scenarios come in matched pairs that change a single feature: whether the action was requested, what is at stake, whether it can be undone, or who will see it. Across ten models from five families, agreement scores mislead in three ways. A simple keyword rule, which we wrote after seeing the benchmark, agrees with our annotators more often than eight of the models, yet its decision changes in only 9 of 48 matched pairs. Equivalent ways of asking the same question change how often a model acts by up to 52.5 percentage points. And every model stops to ask the user less often when it must carry out the task with tools than when it judges a proposed action. When rules are stated explicitly, the same models follow them almost perfectly, so the gaps are not explained by a general inability to follow rules. We release the benchmark and evaluation tools and recommend reporting these properties separately rather than as one score.

Egocentric assistants must connect what users say with what they see across long interaction histories. We formalize this challenge as Spatially grounded Conversational Reasoning (SpaCR): cross-scene, recall-oriented, and counterfactual spatial queries that combine user-stated facts with geometric evidence. Direct vision-language models incur high inference costs and context limits as histories grow, while keyframe selection and retrieval can omit objects or evidence needed for complete recall. We propose Spatially grounded Conversational Memory (SpaC-MEM), an object-centric working memory that uses 3D reconstruction and segmentation to ground conversational information in persistent physical objects. It compresses multimodal histories while preserving spatial evidence and allowing object-specific facts to be updated through dialogue. We also introduce Ego-SpaCR, a benchmark comprising 620 ScanNet video sessions augmented with 95 task-oriented conversations and 3,100 evaluation queries. SpaC-MEM achieves the highest overall answer accuracy among the evaluated methods and improves object recall while requiring substantially fewer reference input tokens than native-video baselines. Removing 3D spatial information substantially degrades performance, highlighting the importance of preserving spatial and conversational evidence together.

The Fréchet distance is a de facto standard for evaluating generative models across domains, appearing as FID for images and FVD for videos. It summarizes the discrepancy between generated and reference distributions in a single scalar, with lower values typically interpreted as better generation quality. However, this scalar view can obscure what drives the comparison. For example, in COCO dataset, increasing the number of diffusion sampling steps improves ImageReward scores yet worsens (increases) FID. Motivated by this mismatch, we seek to make the Fréchet distance more interpretable by uncovering where the discrepancy lies. To this end, we introduce directional Fréchet distance, the expected squared projection of the optimal transport displacement onto a given direction. Across our image, video, and protein case studies, we find that a small number of interpretable directions account for much of the distance. We use these directions to explain the FID increase in terms of semantic concepts represented by CLIP embeddings, quantify FVD's bias toward per-frame appearance, and revisit the interpretation of Protein FID. We open-source our codebase at https://github.com/yhlee-add/directional-fd.

Different low-rank compression methods can produce compressed LLMs that respond differently to the same post-compression recovery procedure, and relative advantages observed between methods at the compression endpoint may shrink, grow, or even reverse after recovery. We ask whether this recovery heterogeneity reflects functional structure beyond scalar loss evolution, and how that structure evolves throughout recovery. Our results establish that this heterogeneity reflects a reproducible compression-induced functional structure, which we formalize as recovery pressure. To characterize this structure consistently throughout recovery, we develop a standardized functional characterization within each backbone that is applicable across heterogeneous low-rank methods. The primary backward characterization reveals reproducible module-wise structure across independent probes, while a complementary forward-only characterization recovers related structure without loss or backpropagation. We further find that recovery pressure measured at the endpoint is associated with subsequent recovery response; during recovery, its module-wise structure is reorganized non-uniformly, and localized updates induce distributed responses beyond directly updated modules. Further evidence indicates that tracking this evolving structure provides a complementary functional view of recovery progress alongside scalar loss.

Large-scale EEG foundation models have demonstrated promising transferability across neurological disorders, but often require millions of parameters and substantial computational resources. In this paper, we present the Universal Semantic EEG Foundation Model (USE-FM), a lightweight EEG foundation model that learns transferable neural representations through self-supervised signal reconstruction on the Temple University Hospital EEG Corpus (TUEG). After pretraining, the encoder is frozen and evaluated on two clinically distinct downstream tasks, abnormal EEG detection (TUAB) and epileptic seizure recognition (TUEP), using a unified frozen-transfer protocol against recent EEG foundation models, including LUNA-Base and CBraMod. With only 1.46 million parameters, approximately one-fifth the size of existing models, USE-FM achieves competitive overall performance, including strong sensitivity and F1-score on TUEP (SEN $75.00 \pm 14.14$, F1 $70.37 \pm 4.01$), while maintaining competitive performance on TUAB (AUC $85.24 \pm 5.61$). Beyond downstream classification, latent representation analysis using $k$-means clustering together with PCA and t-SNE demonstrates that USE-FM learns organized semantic EEG representations comparable to substantially larger foundation models. These results suggest that large-scale self-supervised pretraining enables lightweight architectures to learn transferable semantic EEG representations, providing a computationally efficient foundation for cross-disorder analysis and future clinical decision support in neurological disorders.

In this study, we propose a novel rule-based interpretable choice model, Logic-Logit, designed to effectively learn and explain human choices. Choice models have been widely applied across various domains---such as commercial demand forecasting, recommendation systems, and consumer behavior analysis---typically categorized as parametric, nonparametric, or deep network-based. While recent innovations have favored neural network approaches for their computational power, these flexible models often involve large parameter sets and lack interpretability, limiting their effectiveness in contexts where transparency is essential. Previous empirical evidence shows that individuals usually use heuristic decision rules to form their consideration sets, from which they then choose. These rules are often represented as disjunctions of conjunctions (i.e., OR-of-ANDs). These rules-driven, consider-then-choose decision processes enable people to quickly screen numerous alternatives while reducing cognitive and search costs. Motivated by this insight, our approach leverages logic rules to elucidate human choices, providing a fresh perspective on preference modeling. We introduce a unique combination of column generation techniques and the Frank-Wolfe algorithm to facilitate efficient rule extraction for preference modeling---a process recognized as NP-hard. Our empirical evaluation, conducted on both synthetic datasets and real-world data from commercial and healthcare domains, demonstrates that Logic-Logit significantly outperforms baseline models in terms of interpretability and accuracy.

Geographic information systems increasingly rely on sophisticated spatial representation learning techniques to extract meaningful patterns from complex geospatial data. This paper introduces LEON, a novel self-supervised framework that adapts Graph Masked Autoencoders (GraphMAE) for geospatial region representation learning. Our method leverages the inherent spatial structure of geographic data by constructing hexagonal grid graphs using H3 indexing and applying masked autoencoding techniques to learn robust spatial embeddings from OpenStreetMap (OSM) amenity distribution patterns. We evaluate LEON on multiple real-world datasets including EuroSAT satellite imagery classification and various geographic prediction tasks (housing prices, crime prediction, and urban analytics). Experimental results demonstrate that LEON achieves significant improvements in spatial understanding, with up to 1.87% accuracy improvement on EuroSAT classification and consistent performance gains across geographic prediction benchmarks. The learned embeddings exhibit highly structured and distinct properties, making them particularly suitable for downstream spatial analysis tasks. Our findings suggest that self-supervised learning provides an effective paradigm for geospatial region representation learning using widely available crowdsourced data.

Vision-language-action (VLA) models have shown strong potential as generalist robot policies, but adapting them to unseen tasks often requires costly parameter updates. Recent work such as RICL introduces in-context adaptability by retrieving expert demonstrations based on the current VLA observation and providing them as additional context at test time. The effectiveness of this adaptation therefore depends critically on the retrieval mechanism. In this work, we systematically study how different retrieval methods affect both retrieval quality and task performance within the RICL framework. Specifically, we compare four different methods: image-based retrieval, retrieval augmented with VLA's state, retrieval using features from the VLA backbone, and random retrieval. Our experiments yield three main findings. First, no retrieval method consistently dominates the others in task success, while surprisingly, random retrieval achieves a non-trivial success rate. Second, standard retrieval-quality diagnostics do not reliably reflect downstream VLA performance. Third, demonstrations from different but related tasks can provide useful transferable information. Together, these results provide an initial step toward understanding how retrieval mechanisms shape the in-context learning capability of VLA models and their downstream task performance, while highlighting the need for more careful design and evaluation of retrieval mechanisms for reliable test-time adaptation.

Generative flows sample from an unnormalized target by training a flow to be balanced, and the training loss is the signal a practitioner watches. We ask what that signal is worth: whether a small loss certifies an accurate sampler, whether the loss can be driven to zero, and how fast gradient descent does so. The loss decides the first. Losses that compare the two sides of the balance by their difference bound, in total variation, the error of the sampler the flow implies, with explicit constants that do not involve the policy; flow-matching losses that compare them through a ratio admit no such bound, already on a single cycle, whenever their generator is continuous at balance. On graphs, the backward policy decides the other two. Once it is frozen, balance becomes invariance under the backward chain, so that existence is free on finite graphs, and one constant --- the norm of that chain's Green operator, which plays the role of an inverse spectral gap --- fixes the order of the curvature of the loss around the balanced flow, from above and below, and sets a floor under the rate at which training converges near it. The mechanism is that gradient descent diffuses the flow along the backward policy. For the squared-logarithm generator of detailed and trajectory balance, training the balance loss on states converges globally on every finite path-connected graph, from every positive initialization. The constant can be infinite while backward trajectories are short on average, and exact flow matching can then fail. The bounds and rates are tested by exact computation on enumerable state spaces, and every theorem carries a certification status computed from a Lean~4 development.

Skill Retrieval (SR) aims to identify the most relevant skills from external skill libraries, and becomes increasingly challenging as libraries grow in scale and diversity. Existing methods either rank skills independently or rely on predefined graph propagation and hierarchical routing, making them vulnerable to semantic distractors, local trapping, and early routing errors. We formulate SR as an adaptive information-foraging process that coordinates region-level navigation with skill-level selection according to the utility and uncertainty observed during search. Based on this formulation, we propose SkillGATE, a graph-guided hierarchical retrieval framework with Gate-Aware Monte Carlo Tree Search (MCTS). SkillGATE constructs a graph-preserving hierarchical index and performs adaptive retrieval through selection, expansion, simulation, and backpropagation. G-PUCT guides action selection, expansion explores new regions, simulation evaluates candidate skills, and backpropagation updates search statistics. Experiments on six SR benchmarks show that SkillGATE consistently improves diverse retrieval and reranking backbones, achieving a 16.3\% improvement in overall R@1 over the strongest retriever-based baseline. Our code is available at https://github.com/Edwinbe/SkillGATE-v1/.

Softmax attention gives every token a nonzero weight, which in trained models concentrates into attention sinks and massive activations that widen the dynamic range low-precision inference must cover. Softpick removes this constraint by rectifying scores, eliminating sinks and lowering hidden-state kurtosis, but its advantage fades at scale. We reframe this failure as a normalization problem. Softpick's denominator splits into positive- and negative-shifted sums $D^+$ and $D^-$, used identically in the forward and backward pass, preventing their roles from being isolated. We separate them into a family of operators that independently choose each denominator. The failure originates at initialization: every layer contains rows where $D^+$ is exactly zero, while near-dead rows produce gradient norms above $10^{12}$ regardless of the backward denominator. Only Softpick and a stop-gradient variant, which keeps $D^+ + D^-$ forward but backpropagates through $D^+$ alone, train from scratch. At 230M parameters, the stop-gradient operator matches Softpick on quantization, has fewer dead heads, and retrieves passkeys more reliably, trailing only on peak attention-weight kurtosis.

Recent Test-Time Training (TTT) architectures compress context into fast weights that are updated online and queried as memory. Existing TTT designs keep this memory private to each layer: it recurs only over time, and depth merely indexes L separate memories. We argue that memory ownership need not be tied to depth, and introduce Universal Test-Time Training (uTTT), in which all layers read and write one shared memory while retaining layer-specific backbone parameters. The shared memory thus recurs over two dimensions, time and depth, with chunks and layers as their units: a write by a deep layer in one chunk can be read by a shallow layer in the next. We instantiate this idea as uTTT-MoE and uTTT-Dense. uTTT-MoE routes each token head to a few experts in a pool shared by all layers; uTTT-Dense applies the whole shared memory at every layer without routing. In language modeling, uTTT-MoE reaches 15.5 and 27.9 RULER accuracy at 124M and 760M, 2.6 and 2.1 points above its layer-private counterpart at equal state and active compute, the highest among tested bounded-state models, with per-token loss matching or beating full attention. In novel view synthesis, sharing at fixed per-layer compute gains 0.92 dB in view-23 object PSNR in routed models and 0.76 dB in dense models.

World--action models (WAMs) promise a unified model that predicts action-conditioned futures, generates feasible actions, and supports planning in imagination. However, existing joint video--action models often use computationally heavy, fixed-horizon backbones ill-suited to streaming inference and stable long-horizon open-loop rollouts. We introduce FLEX-WAM, a Flexible and Efficient Block-Causal World--Action Model for unified simulation and policy inference. FLEX-WAM supports variable-length contexts and non-causal prediction horizons, as well as infinite autoregressive generation frame by frame or block by block. Its block-causal, KV-cacheable architecture combines axial attention and blockwise diffusion forcing to enable efficient real-time rollout and deployment-time latency--throughput tradeoffs without retraining. Joint training can nevertheless produce plausible futures that weakly respond to commanded actions. We address this failure mode by balancing state and action flow-matching gradient contributions across the state--action diffusion-noise grid and regulating world-model sampling using Forward-Dynamics (FD) elasticity, an efficient training-time proxy for action responsiveness. Across simulated and real-world datasets, FLEX-WAM achieves superior multi-step prediction quality and latency while producing stable joint state--action rollouts for thousands of steps. As a joint action proposer and simulator within MCTS, it solves long-horizon PushT and all five OGBench Puzzle-4x4 tasks entirely in imagination. On a bimanual OpenArm-based robot, a single checkpoint jointly serves as a play policy and expected-outcome predictor, enabling real-time identification and collection of model--reality mismatches for future self-improvement.

In this pictorial, we consider how the negotiation of creative boundaries with a co-creative AI system can create moments for personal creative reflection. We ground this in our experiences with Froggi-Draw, a single-initiative co-doodling system that gives users power to decide when and how much an AI "collaborator" (Froggi) contributes to their drawing. From a 1-week pilot study where (n=8) novice and experienced artists doodled daily with the system, we report on ways participants navigated creative risk and uncertainty, and how their usage and perceptions of Froggi shifted over time. In moments of disruption, participants described the system as encroaching on their creative territory. We consider how the design of a supportive, co-creative AI "collaborator" might look like the design of a supportive power dynamic---and finding ways to offer users control to find, reflect upon, and flexibly negotiate the boundaries of that dynamic.

Large language models for code generation often fail on execution, multilingual coverage, and contamination control, especially under frozen backbone constraints. We present CodeForge-MA, a unified framework that improves code synthesis through a multi-agent data forge, execution verified reinforced instruction tuning, and a language conditioned mixture of LoRA adapters. Four specialized agents, Composer, Reviewer, Executor, and Curator, iteratively refine instruction code pairs, validate them with tests, and filter duplicates and benchmark leakage. During training, we combine masked supervised fine tuning with a test driven reinforcement objective to align generations with executable correctness. For the larger model, we use sparse expert routing over low rank adapters to improve cross language transfer while keeping the base model unchanged at inference. Experiments show that joint data, objective, and adapter design yields robust gains across programming languages.

Scaling decentralized learning changes not only the number of clients $N$, but also the dynamics of information propagation and consensus. We argue that the effect of increasing $N$ cannot be understood in isolation, because data allocation, topology-dependent mixing, and communication capacity may change simultaneously. We study these coupled effects on CIFAR-10 with $N\in\{10,50,100,200\}$, comparing a degree-two Ring, a Static Random graph, and Local-First Heuristic Evolution (LFHE), a locally adaptive topology process based on friend-of-friend discovery. The Ring provides an analytically transparent failure mode: its Metropolis spectral gap decays as $Θ(N^{-2})$, implying progressively slower contraction of model disagreement as the population grows. Experiments show that holding the nominal local dataset size fixed substantially reduces the apparent population penalty observed when a fixed total dataset is divided among more clients. The remaining degradation depends strongly on communication structure: Ring enters a high-disagreement regime, whereas Static Random and LFHE remain close to consensus. Increasing LFHE's degree threshold further improves accuracy and consensus, but at a substantially higher model-transmission cost. These results show that decentralized scaling is governed by coupled learning and communication dynamics, rather than by the number of clients alone.

Hierarchical reinforcement learning improves long-horizon control by organizing primitive actions around persistent subgoals and assigning credit at multiple temporal scales. Recent hierarchical language agents bring these benefits to interactive tasks by explicitly separating subgoal planning from action execution. We observe, however, that an explicit hierarchy does not by itself determine how stable the resulting temporal abstraction is: the learned boundary policy may replace the subgoal almost every turn, making it effectively transient, or retain a subgoal after it has stopped being appropriate. We call this temporal abstraction instability. We propose Stable Temporal Abstraction via Constrained Optimization (STAC), a constrained boundary-policy optimization method that represents premature replanning and stale persistence as constraint costs. STAC applies the resulting Lagrangian costs only to the sampled boundary decision, leaving the underlying algorithm's rewards, critic targets, subgoal advantages, and primitive-action advantages unchanged. Across two backbones and two benchmarks, STAC improves success over a strong hierarchical baseline by $8.1$ and $7.9$ points on ALFWorld and WebShop with Qwen3-0.6B, and by $23.5$ and $15.8$ points with Llama-3.2-1B-Instruct.

Reasoning errors can propagate into later decisions and memory. This survey synthesizes 312 papers and first-party reports on text-based reasoning hallucinations around three questions: what evidence is observable, what study designs establish, and which corrective actions the evidence supports. UIPCA records unsupported premises (U), invalid inferences (I), dependent reuse (P), visible answer-trace consistency (C), and action-policy failures (A). Across 58 reviewed sources, no comparison establishes that a specified intervention improves reasoning while reducing factual reliability under matched conditions. The synthesis connects diagnosis to verification, repair, selective release, and persistent-state control across memory, tools, and training feedback.

We explain autocorrelation decay power laws omnipresent in texts by self-organized criticality. Specifically, we analyze the recently released KLiCKe keystroke dataset and show that not only the final texts' autocorrelations form a manifold that adheres to a power law with a finite-size scaling, but also the text revisions generate revision-size-dependent restoring dynamics toward that manifold. Thus, human writing appears to dynamically regulate semantic correlations in a text toward a critical state.

Visual search is a fundamental cognitive ability. This study investigates whether Multimodal Large Language Models (MLLMs) exhibit human-like difficulty signatures in visual search tasks. We compared search performance of humans (n = 1,250) and MLLMs using identical 2D and 3D stimuli across different set sizes. Both groups showed efficient performance in feature searches, most clearly when the target had a unique color, but performance degradation in conjunction searches as set sizes increased. Additionally, we found strong correlations between human and MLLM error rates ($ρ= 0.82$), which suggests that MLLMs are sensitive to similar objective complexities, such as stimulus heterogeneity. However, differences were found as well: whereas humans invested extra search time to respond accurately on target-absent trials, MLLMs exhibited extreme present/absent response biases in complex searches. We conclude that MLLMs replicate high-level human performance signatures, yet their underlying computations differ significantly.

Reinforcement learning from human feedback (RLHF) has played a central role in making large language models responsive to human instructions. However, human evaluators often favor flattering or persuasive responses over truthful ones, creating incentives for models to appeal to evaluators at the expense of accuracy. AI safety via debate has been proposed as a way to improve the supervision of language models: in this paradigm, two agents argue opposing positions and challenge each other's claims, potentially exposing falsehoods to the adjudicator. A central premise of AI safety via debate is that truthful arguments are easier to defend than false ones under adversarial scrutiny. In this work, we investigate whether this advantage persists when debaters use rhetorical strategies that exploit biases in human judgment. Inspired by competitive debate, we construct 68 LLM-generated dialogues about detective mysteries with known culprits, spanning four interventions: anchoring, fallacy oversight, pro-jargon, and verbosity. We apply each intervention to either the side advocating for the true culprit or the side advocating for an innocent suspect, allowing us to distinguish influence on adjudication from correctness. In a study with 369 participants, we find that, pooled across bias types, these interventions significantly shift judgments toward the manipulated side. These findings expose a vulnerability in debate-based supervision: human adjudication is sensitive to manipulative rhetorical strategies.

Running the same language model on different graphics processing unit (GPU) vendors can produce different logits, even when the model weights and inputs are the same. We analyze cross-vendor mismatch in two dense and two mixture-of-experts (MoE) models with five metric families, namely bitwise equality, logit differences, top-K consistency, token agreement, and task accuracy. We trace one source of the mismatch to accumulation order inside vendors' matrix instructions. Upcasting to FP32 reduces the dense model's logit error by 43% at three times the runtime, yet keeping only the MLPs in BF16 retains 94% of this gain at 1.3 times the runtime, so most of the cost of full upcasting buys little. In the MoE models, FP32 and FP16 both lower the probability error but raise the logit error and change expert selection, and FP16 fails in the dense model. An output-head low-rank adapter (LoRA) does not help either, since the final hidden state does not predict the mismatch. The mismatch also carries into training. With every seed fixed, a student distilled from a teacher running on AMD answers 431 MMLU questions differently from one distilled from the same teacher on NVIDIA. Under FP32 upcasting, bitwise equality barely changes while the output distributions move most of the way to the reference, so judging cross-vendor agreement by a single measure misreads both its cost and its gains. Code is available at https://github.com/crova-project/crova.

In Networked Markov Decision Processes, transition dynamics are often unknown and the state--action space grows rapidly with the number of agents. In this setting, Taylor representations naturally approximate $Q$-functions, but a naive order-$n$ expansion over $N$ agents requires $Θ(N^n)$ coefficients. We justify these expansions under smooth expected future local rewards with controlled derivatives. Under this condition, finite-speed information propagation and discounting imply that local-critic Taylor coefficients decay exponentially with the graph distance to the farthest agent involved. Discarding distant-agent coefficients and marginalizing then yield scalable local Taylor representations with a bound controlled by graph locality. Building on these representations, we propose a scalable model-free actor--critic algorithm, establishing finite-sample critic and near-stationarity guarantees for a linear LSTD critic. We then introduce a more expressive neural TD parameterization. Unlike prior constructive spectral methods, our approach covers settings without access to a known local dynamics map, such as hidden switched linear--quadratic regulation. Across three control benchmarks, our method matches or outperforms spectral baselines while scaling efficiently to large graphs.

Multimodal models are increasingly shifting toward unified architectures that understand and generate text, images, and other modalities within a shared conversational context. This design enables fluid interaction across modalities, but it also changes the privacy threat model: Information revealed in one part of a conversation may remain accessible when the model later generates content in another modality. This risk is particularly concerning in settings where users rely on locally deployed models for privacy, assuming that sensitive interactions remain confined to their device. We introduce Privacy-Leaking Watermarks (PLWs): invisible, trigger-dependent watermarks that a malicious model provider can condition on prior chat history. With this adversarial intervention, the usual separation breaks: a sensitive keyword or semantic cue mentioned earlier in the conversation can cause a later, unrelated image to carry a hidden yet detectable watermark. PLWs pose a novel threat to users of unified multimodal models: A poisoned model can retain utility while covertly turning image generation into a channel for privacy leakage, even when deployed locally. Across 13 sensitive-attribute triggers and two model families, PLWs reach up to 100.0% TPR at 1% FPR. For example, across all tested conversational separations, OmniGen2 detects every prior disclosure of depression while falsely flagging only 1% of images generated without such a disclosure.

Accurately estimating judgment reliability is a central challenge in adapting LLM judges to newly verified feedback while preserving previously learned behavior. However, existing approaches often rely on output-level confidence, which can be overconfident and poorly aligned with judgment correctness. We propose VERA, a VErdict-conditioned Reliability Axis that estimates reliability from hidden activations by distinguishing correct from incorrect judgments within each predicted-verdict group. Using VERA as a control signal, we develop a VERA-guided periodic adaptation framework that integrates reliability-ranked corrective updates, reliability-residual replay, and periodic refresh of the reliability directions. After VERA-guided adaptation on Chatbot Arena, 8B- and 14B-parameter judges outperform the strongest baseline on each of four held-out public benchmarks, with relative gains of up to 23.01%. The framework also improves focal-class recall by up to 16.1% relative to the strongest adaptive baselines on a separate proprietary temporal auditing task.

Preference-based alignment methods such as reinforcement learning from human feedback (RLHF) and Nash learning from human feedback (NLHF) aggregate pairwise preferences to learn an LLM policy, but a natural goal is maximizing social welfare (average cardinal utility), which comparisons alone need not identify. Gölz, Haghtalab, and Yang (GHY) measure the gap by distortion: the worst-case ratio between the welfare of the best fixed lottery (distribution over responses) and of the learned lottery. They show NLHF is optimal when every user receives the same lottery. Account-based LLMs, however, have information about their users and can serve different lotteries to different people. We give an efficient algorithm, GLHF, that learns a single group-conditioned policy from one comparison per user. Under individual Bradley--Terry comparisons, GLHF asymptotically matches GHY's optimal population distortion bound simultaneously on every group in a prespecified, possibly overlapping collection, with sample complexity growing logarithmically in the number of groups and inversely with the smallest group mass. A sharper guarantee for groups with similar preferences approaches distortion of one when members share a feasible favorite response. In experiments using human coffee ratings and synthetic LLM-generated ratings, GLHF lowers distortion in every evaluated group and substantially reduces worst-group distortion relative to NLHF and other group-agnostic baselines.

Off-policy hierarchical reinforcement learning must estimate the values of high-level decisions while the low-level policy changes. HIRO adapts replay data through subgoal relabeling, but after a label change, the value update targets the relabeled subgoal instead of the subgoal the high-level policy originally needed to update. We propose Hierarchical Time-aware Bootstrapping (HTB), which evaluates specified subgoals under the current low-level policy while retaining accumulated task rewards. Remaining execution time distinguishes subgoal continuation from a new high-level decision. Together with primitive-action conditioning, it enables off-policy Bellman updates based on the stationary environment transition law. HTB combines these one-step updates with multi-step suffix returns and truncated relabeling, reducing dependence on intermediate value estimates. A shared value component supports learning across actions, while nonnegative residuals constrain upward corrections relative to that component. At a fixed mixture weight of 0.95, HTB achieves 32.8% AntFall success versus 9.6% for matched local HIRO over five paired seeds at 10M environment steps. Ablations identify contributions from recursive continuation and mixed supervision; fixed-policy tests show more accurate predictions for actions whose returns were excluded from fitting.

Multiparameter persistent homology is a rapidly developing branch of topological data analysis that improves the robustness of single-parameter persistent homology to outliers, while still capturing the metric characteristics of the data. However, a notable limitation is its lack of scalability. In this paper, we introduce a novel approach for efficiently computing 2-parameter persistent homology on large point sets. Our work extends the Flood filtration, originally developed for single-parameter persistence. Our construction, called the sublevel Flood bifiltration, offers a scalable approximation of the sublevel offset bifiltration. We show that it benefits from theoretical stability properties and describe how to compute it efficiently. We demonstrate the performance of our approach in classification tasks on low-dimensional synthetic datasets, where density awareness is critical, as well as on real-world time series datasets.

Computer-use agents need to capture procedural knowledge of how people use software. User telemetry offers a scalable source of this knowledge. However, learning reusable skills from these logs requires addressing three challenges: (1) Goal Underspecification, since logs do not record the goal behind each action; (2) Non-Replayability, since past activity cannot be replayed to evaluate skill updates; and (3) Interleaved Trajectories, since logs may mix several tasks without marking their boundaries. To address these, we introduce TeleTune, a framework for learning a textual skill library from offline logs without recorded goals, cannot be replayed during optimization, and may interleave tasks. TeleTune uses action-prediction errors on logged trajectories to propose library edits and keep only those that improve held-out action-prediction accuracy, which we call skill-guided progress. The learned workflows also enable retrieval of demonstrations that cover the subgoals of a new task. At test time, the agent is provided with the learned library and the workflow-based retrieved demonstrations. Experiments on WorkArena and Online-Mind2Web show that TeleTune outperforms random retrieval, Agent Workflow Memory (AWM), and their combination. We find that the best baseline varies by setting, whereas TeleTune achieves average success rates of 77.1% and 80.6%, respectively, improving over the strongest baseline on each benchmark by 6.7% and 7.7%. Under the heaviest perturbation of the WorkArena training data,TeleTune keeps the highest average success rate at 68.5%, 6.3% above the strongest baseline. Our analyses show (1) skill optimization and workflow-based retrieval are complementary, (2) optimizing on fixed logs costs 5 to 75 times fewer tokens than validating the same edits with live episodes, (3) skill-guided progress tracks the live success rate.

Molecular optimization must improve target activity and satisfy developability constraints within limited evaluation budgets. Classical methods require tailored rules or training to incorporate chemical instructions and property feedback. Frozen language models can condition edits on this information, but need explicit constraint control and relevant experience. We therefore present PharmAgent, a constraint-aware molecular search method driven by adaptive external state. Its Lagrangian controller translates violations in accepted states into accumulated constraint pressure, keeping this history separate from current property measurements. Structure-indexed replay complements this feedback with relevant evaluated transitions that guide subsequent proposals. As a curriculum progressively activates constraints, candidates and the incumbent are compared under the same current objective, and the accepted state determines the next multiplier update. We derive an exact identity that characterizes how accepted-state violations accumulate in the controller's multipliers. Across five tasks with five independent runs, PharmAgent achieves a summed area under the target-score curves (AUC) of 3.9208 in target-only search, improving over MOLLEO by 37.3%. With online constraints, it achieves a property-adjusted AUC of 0.7076, improving over the strongest online baseline, ExLLM, by 53.8%. These results rank first among all evaluated methods in both target-only and constraint-aware search. The online comparison covers all five baseline frameworks. The full system leads every ablation variant in target quality, property-adjusted performance, and Pareto hypervolume. All five molecular cases reach feasible final states, documenting target gains and trade-offs.

We study learning in continuous games when players receive only pairwise preference feedback, revealing which of two actions is preferred but neither payoff values nor preference magnitudes. We first show that standard external regret and coarse correlated equilibria (CCE) are not identifiable from this ordinal information: the same sequence of play can incur zero and linear regret in two ordinally equivalent games, while the distributions that remain CCE across all cardinal representations consistent with the same preferences are exactly those supported on pure Nash equilibria. Motivated by this gap, we develop a first-order ordinal theory based on normalized unilateral preference directions, introducing an ordinal directional regret benchmark and corresponding equilibrium notions. We show that block-normalized pseudogradient dynamics achieve sublinear ordinal regret and, under additional structure, Nash-convergence guarantees. We then use a single-comparison estimator to implement these dynamics from finite pairwise comparisons. With one comparison per player and round, the resulting algorithm achieves sublinear finite-resolution ordinal regret against arbitrary opponent behavior and, in ordinal potential games, almost-sure last-iterate convergence to the Nash set. Our results provide regret, dynamics, and equilibrium guarantees directly from preference feedback without reconstructing cardinal utilities.

Automated checks of radiology reports may rely on AI-generated checklists that leave findings unmentioned. We used reinforcement learning to train a vision-language model to fill in a 12-finding checklist from a chest radiograph without seeing the sentence under test; a separate checking model judged the sentence from the checklist. On held-out patients, a rule-based check and an independent medical checker, neither used in training, measured discrimination gains (Youden index) of 12.6% and 11.8%; only the rule-based check met the prespecified false-alarm criterion. Switching to the training format, which fixes finding order and enters unmentioned findings as absent, raised the training checker's measured gain and lowered the independent checker's, a prespecified comparison that yielded 6.2% (95% interval 2.0% to 10.5%) and, post hoc on held-out patients, 7.7%. Across 8 checking models, acceptance of a label-consistent negative statement about an unmentioned finding ranged from 1.0% to 97.0%. Labels were report-derived, not radiologist-adjudicated.

Evaluating vision encoders requires metrics that reliably predict their downstream performance in multimodal large language models (MLLMs). Although recent studies have shown that cross-modal metrics can better capture such performance, unimodal metrics remain the dominant choice in practice. In this work, we revisit cross-modal evaluation of vision encoders through large-scale experiments. We identify important limitations in both the experimental design and methodological formulation of prior approaches. After addressing these limitations and introducing simple improvements, we propose RAVEL, a training-free method based on cross-modal nearest-neighbor retrieval. Despite its simplicity, RAVEL achieves state-of-the-art performance across our experiments, outperforming prior methods by a substantial margin. Our results demonstrate that simple cross-modal metrics, when evaluated under a careful and comprehensive setup, can provide a strong basis for evaluating vision encoders for MLLMs.

Action-conditioned time-series forecasting requires accounting for how future actions and exogenous forcings influence multiple targets through partially observed effects with different delays and persistence. Direct conditioning leaves the evolution and target-specific influence of these effects implicit in the predictor, while static relational graphs specify connections without tracking evolving effects. This motivates representing future-driver influence through structured latent states that evolve over the forecast horizon and route information to individual targets. We introduce BeliefGraph-JEPA, a structured latent world model that factorizes driver influence into typed latent-effect states. These states are rolled forward under future drivers and routed through a graph to target-specific nodes, forming the predictive base of a joint-embedding predictive architecture. A capacity-controlled residual supplements this base with direct driver information. On four multi-target clinical, agricultural, environmental, and industrial systems, the framework outperforms a range of pretrained and supervised known-future-covariate baselines. Matched controls isolate latent dynamics, future rollout, graph routing, and residual capacity; future rollout and graph-first residual routing improve forecasting across all four systems.

Text-to-image diffusion models can generate prohibited content, which motivates concept erasure through machine unlearning. Most erasure methods intervene at the text interface, through prompt modification or localized updates to text-conditioning weights, and they are evaluated by what the model outputs for given prompts. Such evaluation cannot see what the network still encodes. Latent-space auditing, which bypasses text conditioning and probes the denoising network directly, shows that erased concepts remain recoverable from internal representations. We find that this also holds for methods built to be robust against adversarial prompts, and that the problem grows with the number of erased concepts. We propose Auditing-Aware Unlearning for Verifiable Concept Erasure in Diffusion Models (AVCE), a framework that grounds erasure in the model's latent representations. AVCE audits the embedding neighborhood of each concept and condenses the discovered vulnerable directions into an anchor at the weakest geometric point. It edits cross-attention and self-attention projections in closed form at this anchor, then fine-tunes the two pathways with pathway-level auditing losses, using orthogonal gradient projection to consolidate multiple concepts. Experiments on SD v1.5, SDXL, and Flux 1.0 across object, explicit-content, and artistic-style unlearning show that AVCE reduces attack success rates by 5.07x and improves auditing scores by 3.84x over the strongest baseline, while preserving competitive generation quality.

A central challenge in continual learning is to acquire new knowledge without forgetting what the model has already learned. This challenge appears in language model training when training data comes from various domain-, user-, or task-specific distributions that are encountered unevenly over time. In such settings, successive minibatches are temporally clustered by distribution instead of being sampled i.i.d. from the overall data mixture. Training on temporally clustered data induces a stability-plasticity tradeoff. Adapting the model to the active distribution can improve the model on the active distribution but may lead to a performance degradation on data it previously trained on. We find that this tradeoff intensifies with longer exposure to the same distribution. We therefore ask if the parameter displacement produced by such a sequence (the task vector) should be fully retained or applied partially. We compare applying the full displacement ($λ=1$) with partial integration, which scales the task vector by $λ$ before applying it to the continuing model and scales the optimizer state by the same coefficient. Across continual pretraining, pretraining from random initialization, supervised post-training, and reinforcement post-training, we find that intermediate values of $λ$ often improve average continuing-model performance relative to full integration, particularly after longer same-distribution sequences. In continual-pretraining experiments with both controlled streams and naturally defined adaptation sequences, task-vector scaling outperforms full integration at the matched learning rate, showing that its benefits are not reproduced by learning-rate scaling alone.

Long-video agents can actively gather question-relevant evidence, but they typically leave a central decision implicit: when has the agent seen enough to answer? We propose CASE, a plug-in termination framework that frames this decision as policy-conditioned sequential stopping. At each causal checkpoint, CASE combines an auxiliary multiple-choice assessment of accumulated evidence with the host agent's execution state. From complete native trajectories, we construct a cost-aware target that compares answering now with stopping later along the same search path, accounting jointly for answer correctness and the full cost of continued reasoning. A lightweight Ridge regressor learns this decision gap and produces STOP/CONTINUE decisions. We evaluate three vision-language models with VideoSeek and AVP. On Video-MME, end-to-end accuracy changes by +0.67 percentage points on average while CASE reduces model-token use by 53.63%. The same frozen policies then transfer zero-shot to LongVideoBench and MLVU, with end-to-end accuracy changes of +3.38 and +4.58 points while saving 58.78% and 51.28% of model tokens, respectively. Across all agent-model-benchmark combinations, CASE attains the highest accuracy-efficiency Pareto-frontier coverage among the compared stopping methods (83.3%) at the selected operating points. Online execution preserves this favorable accuracy-efficiency trade-off and additionally reduces measured runtime by 54.1% on average. CASE provides a plug-in termination framework for long-video reasoning agents, enabling them to decide when further evidence acquisition is no longer worthwhile.

Pull-request review protects software before new code reaches users, helping prevent vulnerabilities that could expose users to attacks. AI agents increasingly perform these reviews and explain which problems need fixing. However, for an attacker submitting vulnerable code, this feedback also reveals what changes may secure approval. Existing PR attacks seek such approval through persuasive text and comments while keeping executable code fixed. This leaves unclear whether an attacker can use the feedback to repair the reported problem while preserving a vulnerability in the revised code. We therefore conduct an empirical study of this threat using AFCRA (Adaptive Feedback-guided Code Revision Attack). To distinguish successful attacks from genuine repairs, we construct AFCRA-Bench from 159 disclosed vulnerabilities, with executable exploits to verify vulnerabilities in code. Across five-round interactions with Sonnet 5 and GPT-5.5 reviewers, AFCRA reaches success rates 2.5x and 12.5x those of the strongest evaluated text- or comment-based attack. Case studies of these successes show how reviewers accept repairs of reported problems while overlooking surviving vulnerabilities. These findings establish feedback-guided code revision as a threat to automated PR review. To address this threat, we derive actionable implications for researchers, AI providers, PR reviewers, and PR authors on securing AI-assisted development.

Autonomous research seeks sustained model improvements through iterative experimentation and feedback. LLM agents show promise in automating machine learning and language-model post-training, but their ability to sustain multimodal improvement remains unclear. We introduce MMPostTrainBench, a benchmark spanning eight tasks in image, audio, video, and joint audio-video understanding and image-grounded software repair. Agents operate from a common base model within fixed budgets, using development feedback before independent evaluation of their submitted models. Evaluation covers target and non-target model outcomes, iterative model improvement and selection, and research integrity. Across all eight tasks, 52.1% of model--task means fall below the base, and evaluated submissions also exhibit non-target regressions. Model performance does not consistently improve across research iterations, and agents do not reliably select the best evaluated candidate for submission; final submissions trail that candidate by up to 5.38 percentage points. Extending autonomous research from text-only to multimodal tasks introduces additional sources of error in perception, cross-modal alignment, and temporal grounding. The observed regressions and selection gaps highlight the need to balance targeted improvements with non-target capability preservation and to retain gains across research iterations. These requirements motivate MMResearch, a multimodal research framework that connects media-grounded evidence to hypotheses and interventions, carries findings across rounds through hierarchical memory, and retains candidates using development evaluation. Added to existing code-agent runtimes, it improves submitted-model accuracy by up to 7.75 percentage points for Claude Opus 4.8 with Claude Code and 2.33 points for GPT-5.6-sol with Codex.

Generative modeling of graph-structured data is crucial for tasks ranging from drug discovery to social network simulation. Among these models, denoising diffusion models have achieved great success in graph generation by learning to progressively reverse a process that adds noise to the original graph. However, the standard noise-prediction approach of diffusion models is suboptimal for graph data. The goal for a graph generative model is to learn the clean graphs' topological properties, such as connectivity and degree distribution. Because a diffusion model that predicts noise does not explicitly learn these topological properties, it is challenging for the model to output graphs with the desired structural statistics. To address this challenge, we introduce Direct Graph Flow Matching (DiGFM), a novel graph transformer model guided by two goals: predict clean graphs and improve sampling efficiency. Distinct from the prevailing diffusion approach, DiGFM employs a continuous flow-matching paradigm and integrates direct graph prediction. Specifically, DiGFM maps the prior noise distribution to the clean graph distribution via a multi-step process: the model repeatedly predicts the underlying clean graph, and a transformation is employed to convert the model output to the velocity vector that points in the direction toward the clean graph distribution. This design enables DiGFM to generate high-quality samples using only 2.5% to 15.6% of the steps required by diffusion-based models, which leads to a 5.3x to 257x speedup in wall-clock inference time. Experiments demonstrate that DiGFM outperforms or matches prior state-of-the-art models across general graph benchmarks and molecular datasets, generating graphs with strong adherence to ground-truth structural statistics at significantly faster inference speeds.

Conversational database interfaces face a critical challenge: users naturally embed queries in conversational noise (greetings, politeness, off-topic remarks), which degrades intent classification accuracy and wastes computational resources. Despite advances in orchestration and retrieval strategies, a fundamental question remains unanswered: which pooling strategy maximizes intent classification accuracy under realistic conversational noise in production language models? This work addresses this gap through 360 controlled experiments spanning four pooling configurations (mean, max, last-token, attention, and FFT-augmented variants) using Llama-3.2-1B-Instruct on BANKING77 and CLINC150 datasets under clean/noisy conditions with ten random seeds. Key findings reveal that attention pooling consistently outperforms alternative strategies under noisy conditions (~+2.6-2.8 F1 over the default), while mean pooling degrades performance by up to ~5 F1 points. Frequency-domain filtering does not produce consistent accuracy improvements and functions primarily as a structural variation rather than an accuracy-enhancing component. These results provide concrete, evidence-based guidance for building noise-robust conversational classifiers: attention pooling is recommended for noisy interfaces, mean pooling should be avoided, and last-token pooling is appropriate for clean-query scenarios.

Large language models (LLMs) have demonstrated strong performance on coding and reasoning benchmarks; however, their ability to solve graph-structured machine learning problems remains largely unexplored. In particular, no benchmark currently evaluates whether LLMs can autonomously solve end-to-end Graph Neural Network (GNN) coding tasks under realistic competition settings. To address this gap, this paper introduces GNN-CB, the first competition-based benchmark for evaluating both humans and LLMs on GNN coding tasks. GNN-CB consists of 18 curated competitions spanning node-, edge-, and graph-level prediction across diverse graph categories, domains, and difficulty tiers. All submissions are evaluated through a unified automated pipeline with hidden test sets and standardized scoring. Human participants solve tasks under controlled competition constraints, while LLMs are evaluated using a frozen zero-shot prompting protocol based on a plan-then-code paradigm with bounded execute-and-repair loops. The benchmark additionally supports both non-agent and autonomous agent-based evaluation within the same protocol. Under our evaluated protocol, LLMs rarely match Human Top performance and show less stable performance across competitions. No single model dominates: a few competitions are won by LLMs, yet humans still hold the top score on most tasks. We release GNN-CB as a living benchmark with automated evaluation infrastructure, dynamic leaderboards, and reproducible execution pipelines. Beyond benchmarking, GNN-CB provides a practice-oriented resource for studying GNN implementation across progressively diverse graph-learning tasks. The benchmark and evaluation framework are publicly available at https://basiralab.github.io/GNN-CB/.

Small language models (SLMs) offer a promising foundation for on-device agents through low-latency, resource-efficient inference, yet limited reasoning and planning capabilities constrain their performance on long-horizon tasks requiring multi-step interaction with the environment. Step-level collaboration between SLMs and larger cloud-hosted models can bridge this gap, but identifying states that warrant cloud assistance remains challenging: the contribution of each cloud call is entangled with subsequent actions and can be assessed only from the final task outcome. Compounding this challenge, the SLM must balance two competing objectives: maximizing task success and minimizing cloud calls. To address this, we propose Sibyl, an algorithm that trains SLM agents to selectively consult cloud models at the step level and internalize their guidance for subsequent decisions, achieving strong task performance with minimal cloud reliance. Sibyl follows a three-stage training pipeline that (1) builds a robust base policy through consultation-free self-evolving reinforcement learning (RL); (2) cold-starts consultation behavior via decisive-disagreement state mining; and (3) jointly optimizes consultation decisions and guidance internalization through consultation-aware RL. Experiments on ALFWorld and WebShop demonstrate that Sibyl, using only a 0.6B-parameter model, outperforms state-of-the-art baselines, including agent training and routing methods, by 95.2% and 80.4% in success rate while averaging only 0.8 and 3.9 cloud calls per trajectory, respectively.

Long-horizon search requires agents to gather evidence across multiple steps and synthesize it into well-supported answers. The recent agent harnesses provide a natural and promising framework to support such long-running search processes. As interaction histories grow, one single agent in harnesses might get stuck and cause the policy to lose track of unresolved questions, overlook useful evidence, or terminate before sufficient support has been collected. One of promising way is to decouple three distinct responsibilities of proposing retrieval actions, updating persistent state, and deciding when to stop rather than concentrating them within a single policy. Targeted at it, we introduce Harness-Search, a multi-agent search harness to reduce the local errors propagating across subsequent exploration, evidence curation, and termination decisions. In particular, Harness-Search assigns these responsibilities to three permission-bounded authorities: a Retrieval Policy that proposes search operations, a Memory Operator that validates and commits persistent-state updates, and a Summary Auditor that accepts or rejects termination based on the sufficiency of the curated evidence. Together, these roles form a Propose-Commit-Audit loop in which actions are proposed, persistent evidence is selectively committed, and stopping decisions are subjected to an explicit sufficiency check. Across seven long-horizon search benchmarks, Harness-Search improves both retrieval and answer generation under the same policy backbone, increasing Recall by 4.60-27.92 points and Final-Answer Recall by 12.34-30.13 points over the strongest harness-based baseline on each evidence-retrieval benchmark. Moreover, trajectory-level analyses show that Harness-Search continues to accumulate useful evidence and expand evidence coverage with less redundant retrieval as the search history grows.

Large Language Models (LLMs) are increasingly applied in hardware design automation, demonstrating strong potential in generating and understanding hardware description languages. However, most existing benchmarks focus on Verilog, with limited evaluation of VHDL, which remains widely used in industry and academia for FPGA and safety-critical systems. To address this gap, we introduce VHDL-REPOBENCH, a large-scale, cross-file, repository-level benchmark for assessing LLM capabilities on realistic VHDL design generation and analysis tasks. VHDL-REPOBENCH curates ~100 open-source VHDL repositories, encompassing ~2.5k VHDL files and ~500 testbenches, and provides structured problem statements, module stubs, and self-verifying testbenches. The benchmark enables comprehensive evaluation across syntax, semantic correctness, hierarchical reasoning, cross-file dependency resolution, and functional verification. We evaluate several state-of-the-art models, including GPT-4o, Llama-3-70B, Qwen2.5-72B, CodeLlama-70B, and multi-step reasoning approaches such as Reflexion and CoDes. Results reveal that while current LLMs achieve moderate line- and block-level accuracy, substantial challenges remain in multi-file reasoning, hierarchical design understanding, and specification-to-module generation. VHDL-REPOBENCH represents the first large-scale VHDL-focused benchmark and provides a valuable resource for the hardware design community to evaluate, compare, and advance LLM capabilities for practical VHDL development.

Model performance monitoring is a standard practice in machine learning deployments. Detection performance is tracked continuously, and model decay is expected as the relationship between the feature and target variables degrades, a phenomenon known as concept drift. Explanation fidelity, however, is rarely monitored with the same discipline, even in domains such as financial systems, healthcare, and other regulated environments where explanations are required for governance purposes. This paper investigates whether explanations can decay under data drift, even when the feature-target relationship remains stable, and whether explanations produced before drift occurs remain faithful to the decisions of the model that replaces them. Using the IEEE-CIS Transaction Fraud Detection dataset, we find statistically significant covariate shift but no statistically significant evidence of concept drift under the implemented conditional-drift tests, thereby providing an empirical setting in which input distributional change can be studied separately from detectable changes in the feature-target relationship. Local explanations are generated with ExIFFI and evaluated at three levels: path validity, structural behaviour, and fidelity under controlled intervention. Results show that while prior explanations retain substantial decision relevance to a retrained model, they are consistently less faithful than newly generated explanations, with no evidence of a systematically widening gap across the evaluated windows. The study shows that explanation fidelity requires its own monitoring, that structural stability of explanations does not guarantee functional fidelity, and that explanations should be treated as artifacts tied to the model that produced them.

Collaborative learning typically exchanges model parameters: federated clients communicate updates, while independently adapted foundation models are combined by exchanging adapters or checkpoints. This makes communication scale with model size and requires local specializations to be reconciled in weight space, where interference is common. We ask whether knowledge can instead be shared through model behavior on task-unrelated inputs. We introduce Distributed Subliminal Learning (DSL), a collaborative learning primitive in which participants adapt a common model locally, probe it with task-unrelated inputs, and transmit only the resulting carrier outputs. A coordinator pools these outputs and distills them into a shared model. The primitive supports one-shot foundation-model composition through carrier completions and iterative federated learning through carrier logits, without transmitting model updates or requiring task-related proxy data. In LLM composition, compared with LoRA averaging, DSL achieves higher preference retention (94.56% vs. 87.76%) and a larger GSM8K gain over the base model (22.0 vs. 0.6 points), while reducing upload by 30.6-49.0$\times$. In federated classification, DSL reaches 96.83% on MNIST with 8.9$\times$ less uplink than FedAvg and provides lower-communication operating points on CIFAR-10 and Tiny ImageNet. These results establish random-carrier outputs as a practical communication primitive for knowledge sharing across distinct collaborative learning paradigms.

Reconstructing physical fields from training samples that are always incomplete requires learning spatial structure from fragmented observations.Existing context--query work establishes how held-out observations provide valid training targets, but this does not make the complete-field distribution identifiable when every training field is incomplete.With finite data, weak evidence of sharp transitions and localized variations can further favor averaged predictions that attenuate local detail.A structural prior is therefore needed to favor plausible completions; local spatial relationships offer one grounded in the observations.We propose a locally constructed, globally revisable estimator that explicitly learns local field estimates and subsequently corrects them using full-domain observations.A shared coordinate-conditioned predictor learns from incomplete patches, allowing relatively well-observed neighborhoods to provide direct supervision of local structure.Its overlapping predictions are reconciled into an observation-conditioned consensus field.A full-domain estimator retains the original observations and learns a residual correction around this frozen field estimate, allowing locally constructed structure to be revised by broader evidence.The local estimate serves as both an explicit input, accompanied by its discrepancies with the observations, and a prediction starting point that the global model can revise.On three real-world ocean datasets with authentic observation gaps, our estimator achieves the lowest MSE and highest PSNR on withheld source-supported values, reducing MSE by 28.9\%--34.5\% against the strongest external baseline.

On-policy distillation (OPD) has emerged as an effective post-training paradigm for language models, with recent efforts extending it to block diffusion language models (BDLMs). However, existing studies focus almost exclusively on small block sizes, leaving distillation into student models with larger blocks underexplored. In this work, we investigate this regime and reveal two critical optimization biases that induce severe training instability. First, mismatched block boundaries between teacher and student cause \textbf{\textit{context misalignment}}, providing distorted supervisory signals that misguide student decoding. Second, even under aligned contexts, an \textbf{\textit{intrinsic optimization bias}} in OPD, where the student tends to rapidly absorb high-support signals while lagging on low-support updates, drives a premature confidence surge that traps weaker students in catastrophic overconfidence collapse. To resolve these, we propose \mbox{\textbf{Un-OPD}}, an unbiased on-policy distillation framework with two novelties for stabilizing BDLM training. First, Un-OPD introduces a boundary-aware step filtering strategy that eliminates context-misaligned decoding steps. Second, Un-OPD proposes moderating optimization intensity at high-support positions via a support-rebalanced confidence calibration, thereby bypassing overconfidence collapse. Beyond stability, we also introduce a rollout reuse mechanism to reduce rollout generation overhead. Extensive experiments on math reasoning and code generation benchmarks show that Un-OPD consistently stabilizes training and delivers superior performance while reducing wall-clock training time by approximately half.

Ensemble guidance combines pretrained diffusion priors with black-box forward models to solve inverse problems without differentiating through the physical simulator. However, observation coordinates with large predictive spread or extreme residuals can dominate the ensemble correction, degrading reconstruction accuracy. We show that two simple modifications, weighting and clipping, substantially improve this correction. Our method, Robust Ensemble Guidance (REG), uses ensemble predictive spread to balance observation scales and adaptively clips standardized residuals to limit the influence of extreme discrepancies. Both operations reuse existing particles and forward predictions, requiring no additional denoiser or forward-model evaluations. Under a local linear Gaussian model, we derive conditions for reduced one-step estimation risk, bound the influence of individual observation coordinates, and characterize when these benefits persist with finite ensembles. Experiments on Navier-Stokes inversion, black-hole imaging, and acoustic full-waveform inversion demonstrate improved reconstruction over the underlying ensemble solver. In particular, REG increases black-hole reconstruction PSNR by 6.2-8.2 dB across three observation regimes and reduces Navier-Stokes reconstruction error by 26.4\% in a matched-budget comparison. These findings highlight the importance of observation heterogeneity and residual influence in designing reliable generative solvers for scientific inverse problems.