The Inference Report

This paper examines the role of Novel View Synthesis (NVS) in geometric representation learning. In principle, NVS should reason about 3D scene structure, thereby enabling transferable multi-view geometric representations. Yet, existing encoder-based NVS methods yield poor representations. This is not because of a lack of supervisory signal, but rather due to inconspicuous architectural choices: \textit{spatially expressive decoders} that dilute representational capabilities of the scene encoder, and \textit{low-level pixel-space targets} that hinder feature learning. We present SNAP, a self-supervised encoder-decoder transformer that addresses both through a pose-conditioned local decoder and a latent-space reconstruction objective. SNAP is task agnostic, and we show that it is competitive with special-purpose geometry-supervised methods. SNAP also performs competitively against self-supervised representations across five tasks: visual localization, pose estimation, point correspondence, depth estimation, and robot manipulation. Remarkably, SNAP's patch features exhibit emergent viewpoint invariance that approaches heavily supervised models despite lower compute and data budgets. Under camera shifts where standard 2D representations collapse, SNAP degrades more gracefully, revealing that restricting decoder expressivity actively prevents the suppression of transferable geometric structure. https://snap-nvs.github.io

We introduce 4DCodeBench, a benchmark for 4D inverse graphics through code generation, in which agents reconstruct dynamic scenes from video as executable graphics programs. To accomplish this, agents must translate visual observations into compact representations of scene structure and dynamics, by implementing abstractions such as physical simulations to reproduce complex behavior. To evaluate this capability, we curate a set of real-world videos and construct synthetic scenes spanning diverse physical phenomena, including deformation, fluid flow, and fracture. We perform extensive benchmarking of frontier models, finding that strong static reconstruction capabilities do not yet translate into reliable reconstruction of complex dynamics. 4DCodeBench provides a testbed for tracking progress toward agents that can interpret the dynamics of the world through code. Our benchmark is available at https://github.com/4DCodeBench/4DCodeBench

Continual learning treats degradation on previously seen data as evidence of failure, a convention inherited from settings with a stationary prediction target, where a correct label remains correct indefinitely. World models do not satisfy this condition. Their prediction target is the environment, which changes, so knowledge that was accurate when acquired may later become false, and discarding it is required behavior rather than a defect. Non-stationary ground truth is well studied in the concept drift literature and in the temporal factuality of language models, but has not been formulated for world models, which are distinctive in that they also encode knowledge that must never be revised. We argue that continual world models require retention stratified by invariance timescale, separating invariants such as physics and object permanence, which must never be revised, from instance-level facts that should be revised as soon as the environment changes. Standard forgetting metrics cannot distinguish a world model that has correctly revised outdated knowledge from one that has suffered catastrophic forgetting, and consequently rank a frozen model highest, while existing physical-reasoning benchmarks evaluate only frozen checkpoints. We propose differential retention, which reports invariant regression testing across the adaptation stream jointly with revision latency, without aggregation.

Ribonucleic acid (RNA) functions through conformational changes that are not fully captured by static structures. However, large-scale standardized RNA dynamics data remain limited, and existing approaches typically treat trajectory generation and dynamics understanding as separate objectives. Here, we introduce RNADynBench, a standardized RNA molecular dynamics (MD) benchmark with 2585 quality-controlled 100-ns all-atom trajectories and leakage-controlled splits. Building on RNADynBench, we develop RNADynNet, a unified model for RNA dynamics learning that uses a shared backbone for both trajectory generation and dynamics fingerprint extraction from a single conformer. It combines coordinate denoising, single-frame-to-trajectory alignment, and physical grounding to connect all-atom trajectory generation with dynamics representation learning. Physical grounding improves both generated dynamics and the physical information recoverable from these fingerprints. Across both test sets, including the high-flexibility challenge set, the generated trajectories achieve RMSF correlations of 0.875 and 0.766, while single-conformer predictions show comparable agreement with MD-derived dynamics. RNADynBench and RNADynNet together establish a benchmark and unified modeling framework for generating and understanding RNA dynamics.

Inspired by human vision, we introduce a framework using active gaze to enable fine-grained bimanual manipulation with only a single stereo camera. EyeRobot 2.0 physically attends to a 3D fixation point in the scene by swiveling two eye viewpoints to center their gaze on it. The resulting images are processed foveally by allocating more visual tokens to the image centers, focusing computation on task-relevant features. Such Active Visual Fixation (AVF) requires carefully coordinated gaze during task execution, which we accomplish hierarchically by first training a low-level gaze servoing policy conditioned on a goal object, then training a target selector which emits fixation goals based on task progress. Both modules are trained with RL on real-world data: the first is trained with a dense geometric reward and the second co-trains with the BC gripper policy which allows it to discover fixation sequences that can resemble a human's fixation sequence while performing the task. EyeRobot 2.0 further takes advantage of fixation by canonicalizing gripper information into a fixation-relative SE(3) frame, which compacts the size of the action distribution to learn. We collect teleoperation data for 7 real-world and 6 simulated tasks, and conduct over 1000 physical and 1800 simulated robot trials comparing EyeRobot 2.0 against passive stereo and ego + wrist camera policies trained on the same data. Removing wrist cameras is costly for standard policies: with only passive stereo, real-world success drops from 52% to 27%. EyeRobot 2.0 closes this gap with only stereo, outperforming passive stereo by 40% in real and 20% in sim. It matches ego + wrist policies when their wrist views are clear (69% vs. 64%), and more than doubles their success when grasped objects occlude the wrist cameras (48% vs. 22%)

We study the minimization of sums of smooth strongly convex functions over undirected graphs, with each function held by one agent and communication restricted to neighbors in the graph. Existing decentralized methods, whether based on gossip or on routing over spanning trees, typically use the network to mix or aggregate information to enable {\it prescribed} local optimization updates. What this communication-centered viewpoint lacks is a general framework that uses graph structure to {\it jointly} design the optimization subproblems and the cooperative computation and communication through which agents solve them cooperatively. We develop such a framework from first principles, jointly designing the linear representation of agreement constraints, the blocks of the resulting dual variables (jointly optimized), and connected cluster of agents that cooperatively solve each block subproblem over the assigned subgraph. GATE (Graph-Tearing message passing) is a first instance of this framework: one variable per edge and tree blocks. At each iteration, agents update their assigned edge variables by minimizing the sum of the two endpoint cost-to-go messages and relaxing the result. The messages are updated through local minimizations following the tree recursion. To reduce per-iteration computational and communication costs, we develop GATE-S, a surrogate variant using tractable local models and lightweight message parametrizations. We establish linear convergence with a rate explicit in the interplay among function regularity, network topology, and the chosen partition, revealing the effects of graph decomposition. Numerical experiments are conducted to validate the theoretical results and evaluate the efficiency of our algorithms.

The choice of post-training data for large language models substantially affects downstream performance. Gradient-based data selection is a popular approach that ranks training data by how well their gradients align with those of a small validation set. However, ranking with full-parameter gradients requires an expensive backward pass on every sample, making computation intractable for large candidate pools. This raises a natural question: can we approximate full-gradient features at a fraction of the cost? Conveniently, we find that output-layer gradients suffice for effective data selection, yet require only the cheaper forward pass. We implement this as LESSER, a drop-in wrapper for selection methods that reduces the feature-extraction FLOP cost by $9.7\times$ for SFT and $3.0\times$ for RL benchmarks, while tracking full-gradient performance on downstream tasks. Empirically, we find that even when output-layer and full gradients rank individual samples differently, they select batches with aligned gradients.

Modern chess engines are silent experts: they play at a superhuman level, but do not offer explanations for their play. On the other hand, language models (LMs) can generate plausible-sounding explanations, but their weak playing strength limits the utility of their explanations. We introduce Queen, a 4B-parameter chess-language model that can explain its moves and plans while playing at the level of a typical Grandmaster. Our novel framework enables domain-specific reasoning through complementary components: an encoder-decoder architecture and an iterative distillation algorithm. This architecture integrates a silent expert chess encoder with an instruction-tuned LM through cross-attention, which we train via a question-answering curriculum to extract chess concepts from the encoder's representations. Building on this domain-adapted model, we iteratively improve its explanations with a natural-language analog of the Bellman update: the model analyzes the positions after its top candidate moves and consolidates them into an explanation of the current position, which is then distilled back into the model. Over seven iterations, our model gains over 900 Elo points (1782 to 2697), substantially surpassing all frontier models on both playing strength and puzzle accuracy, despite containing three orders of magnitude fewer parameters. Furthermore, LM-based evaluations show that our explanations are fluent and approach GPT-5.6-Sol (high) in coherence. The generality of our architecture and training procedure suggests a recipe for applying language models to domains where silent expert encoders are available, like games, robotics, and computer use.

Scarcity of labeled data limits development of deep learning biomarkers in oncology. We develop a two-stage AI model predicting pathological complete response (pCR) to neoadjuvant therapy in breast cancer. The first stage learns the transcriptome from histopathology using 8,742 patients across 32 cancer types, corroborated by pathologist review and spatial agreement with measured expression. This simplifies the second stage to predicting pCR from inferred expression and clinical variables. Developed using 1,080 patients (five cohorts) and evaluated in 1,412 patients (nine cohorts), the model achieves a pooled AUROC of 0.79 (95% CI, 0.73-0.85), discriminating responders within molecular subtypes. It outperforms histopathological biomarkers, remaining stable across intratumoral sampling and with minimal biopsy tissue. Ablations show transcriptome-wide inference improves discrimination over clinical variables alone or one-stage pathology models, and robustness by avoiding genomic assays' gene selection constraints. These results indicate that biologically informed compression may generalize to data-sparse applications in precision oncology.

The dynamics of cells, organisms, and fluids are often modeled as probability distributions evolving over time. Reconstructing and extrapolating this evolution from unpaired snapshots requires assumptions about the underlying process. Wasserstein gradient flows are a common choice, but they cannot describe conservative or periodic dynamics. Lagrangian mechanics in Wasserstein space covers both, but existing methods for learning it are simulation-based: they run a numerical solver at every training step, which makes training expensive. We propose Double-Stitch, a simulation-free method that learns these mechanics by penalizing the residual of the equation of motion along a learned population path. We derive this equation from a Clebsch variational principle that does not require gradient velocities, and show that the residual vanishes exactly when the equation holds. We test Double-Stitch on synthetic, single-cell and ocean vortex datasets and find that it matches or outperforms gradient-flow methods and simulation-based WLM on most tasks, while training $4$-$14$ times faster than WLM. We provide a JAX implementation of Double-Stitch at https://github.com/BasisResearch/stitching.

LLM-guided evolutionary methods, such as AlphaEvolve, have emerged as powerful approaches for challenging computational optimization problems, such as circle packing. However, prior work typically optimizes performance gain over a fixed number of iterations. We argue that practical optimization should maximize gain per unit cost. To this end, we propose FrugalEvo, a cost-aware evolutionary framework where a stronger, higher-cost LLM explores solution strategies, and a cheaper LLM implements them and iteratively refines the resulting code. We also design a cache-efficient evolution process, where our harness and prompts maximize the sharing of prefixes across different evolution steps, to improve cache reuse. To measure solution quality throughout a fixed cost budget, we introduce Budget-Aware Area Under the Curve (BA-AUC), defined as the area under the best-so-far evaluation score curve over cumulative LLM cost, up to the budget. Across 10 mathematical and systems optimization tasks, FrugalEvo matches or surpasses state-of-the-art baselines, including OpenEvolve, ShinkaEvolve, AdaEvolve, and EvoX, in final solution quality and achieves higher BA-AUC on 9 tasks. It also achieves higher average performance than these baselines on 10 algorithmic optimization tasks from ALE-Bench-Lite. Notably, on circle packing, FrugalEvo achieves new state-of-the-art performance with GPT-5.6 Terra and Luna for only 1.68 USD and with GLM-5.3 and its Flash variant for only 0.55 USD, matching or surpassing all baselines, including multi-agent methods such as CORAL and SwarmResearch, which cost approximately 50 USD on average.

Policy-gradient methods are central to modern reinforcement learning, including LLM post-training. When they struggle, the usual suspects are exploration, credit assignment and action-sampling noise. Classification has none of them. A classifier is a policy whose expected reward, its \emph{expected accuracy}, is the probability it assigns to the correct label, and because that label is known, the policy gradient is exact and smooth. Yet exact policy gradient loses to cross-entropy, even on expected accuracy. The exact gradient is myopic: it values an update only by what it buys now, but each update also sets where the next one starts, so an update's value depends on how much learning remains. Viewed this way, cross-entropy is patient accuracy, the total error an example would pay if its log-odds rose at unit speed forever, while exact policy gradient is the zero-horizon limit. Truncating this total at the learning that remains yields the horizon loss, a one-line change that moves from cross-entropy toward exact policy gradient as training runs out. In a simple allocation model, it provably escapes the trap that catches each endpoint. On MNIST and on ImageNet with ResNet-50, ResNet-101 and ViT-S/16, the horizon loss improves top-1 accuracy over cross-entropy at a flat learning rate, and the gain grows with label noise.

Masked diffusion language models (dLMs) offer a promising parallel alternative to autoregressive models for complex reasoning. However, they face a distinct credit-assignment challenge, since a few commitments during denoising sharply reduce the uncertainty over the remaining masked positions and shape much of the response. Most post-training recipes for dLMs do not use this signal to decide which tokens to train on: they typically train on the final text or assign rewards to whole denoising steps, rather than selecting the individual commitments that shape the response. We introduce Pivot-SD, an efficient offline self-distillation framework that supervises only these high-impact commitments (pivots). Pivot-SD selects pivots using an information-gain metric measuring uncertainty reduction over the remaining masked positions. Pivots from successful trajectories are trained with cross-entropy, and pivots from failed trajectories with targeted unlikelihood, leaving the rest of the failed trajectory untouched. Using only 200 questions and four rollouts each, Pivot-SD improves LLaDA-8B-Instruct over full-sequence SFT and budget-matched diffusion RL baselines across math and code benchmarks.

Deploying a new decision policy creates a cold-start problem for prediction models whose targets depend on the policy's actions: historical observations reflect earlier policies, while real observations under the new policy are not yet available. Simulation offers a way to address this gap by rolling out the target policy across counterfactual scenarios and using the resulting trajectories to learn how the system responds to those controls. The simulation-to-reality (Sim2Real) transfer of this simulator-trained model can then be backtested by evaluating it against real observations from past deployments. Using two real-world inventory-control deployments, we evaluate this process from three angles: simulator fidelity, zero-shot transfer to real behavior, and adaptation as real target-policy observations accumulate. The simulator-trained forecaster achieves lower point-estimate mean absolute percentage error (MAPE) than the same architecture trained on historical real data, reducing MAPE by 1.2-3.1 percentage points in Study 1 and 12.5-18.7 points in Study 2. After deployment, lightweight calibration using early real observations further reduces error by up to 2.5 percentage points. These results provide empirical evidence that simulator-generated counterfactual data can support cold-start forecasting under a new policy, and the resulting model can be further refined as real deployment data become available.

Multi-pitch estimation in vocal ensembles is challenging because singers occupy overlapping pitch ranges and often sing at closely spaced fundamental frequencies, causing their harmonics to overlap in time-frequency representations. Existing models commonly use harmonic constant-Q transform (HCQT)-based representations to provide frequency-adaptive resolution, at the cost of expensive feature extraction when training mixtures are generated on the fly. We revisit this design and compare HCQT with a linear short-time Fourier transform (STFT), whose frequency bins are directly provided as model inputs. Despite its fixed frequency resolution and the absence of a pitch-aligned input grid, the linear STFT outperforms HCQT while substantially reducing feature-extraction cost. Further analysis shows that a longer analysis window or broader spectral coverage provides no additional improvement, while restricting the input to the predicted pitch range reduces the advantage of the linear STFT. These results suggest that finer frequency resolution does not necessarily improve vocal-ensemble MPE, and that shorter analysis windows can be more effective for time-varying vocal pitches.

Dense-retrieval services must switch among embedding-prefix dimensions and index bit rates as latency, quality, and memory budgets change. Tuning a quantizer separately for each rate gives the best quality, but the retrieval tier then holds several code streams and quantizer states at once. We introduce Matryoshka Residual Vector Quantization (MRVQ), a post-hoc residual quantizer for frozen embeddings. Its maximum-rate code can be truncated two ways: dropping residual stages lowers the rate, and dropping embedding coordinates lowers the dimension. One resident artifact therefore serves every (dimension, rate) pair we evaluate. Across FiQA and NFCorpus, four embedding families, and {4, 8, 16}-byte codes, MRVQ is the lowest-RAM design we evaluate. It uses 17.8-22.0x less memory than three separately trained QINCo2 indices, and 1.89-2.02x less than a lean shared-model steelman. The saving is not free: per-rate QINCo2 is 0.026-0.107 nDCG@10 better on FiQA. But MRVQ beats PQ, OPQ, and AdANNS-OPQ at matched code size. We also evaluate a low-build-cost PCA-scalar design that attains quality comparable to RaBitQ and its extension while fitting 420x faster at the median. Finally, we report two negative results: QINCo2 collapses when trained at high rates, and a ranking-bound hypothesis misses its pre-specified acceptance criteria. MRVQ is therefore a low-memory operating point for elastic retrieval, not a universal quality winner.

Federated Learning (FL) is a privacy-oriented learning paradigm that enables collaborative model training while keeping training data local to participating clients. However, it does not guarantee that clients submit policy-compliant contributions or that aggregators process admitted contributions correctly. Existing verifiable FL systems tailor validation rules to specific FL settings, learning workflows, and cryptographic constructions, limiting their applicability across network topologies, participant roles, and aggregation semantics. In this paper, we present PoCoFL, a policy-compliant federated learning framework that separates three aspects: (i) FL type, (ii) policy semantics, and (iii) cryptographic realisation. We provide a formalisation that captures client and aggregation requirements as policy-dependent relations. Clients prove compliance of their contributions using commitments and non-interactive zero-knowledge proofs, while aggregators prove that the recorded set of admitted contributions was processed according to the selected aggregation policy. We demonstrate PoCoFL through four formal instantiations: (i) vanilla, (ii) continual, (iii) personalised, and (iv) threshold-encrypted federated learning. We evaluate the effects of policy enforcement on the learning objectives of vanilla, personalised, and continual FL. We further implement proof-of-concept realisations of all four instantiations, demonstrating the versatility and practical feasibility of PoCoFL. Overall, these results show that PoCoFL can capture complex policy representations while remaining network-topology agnostic.

Marine ecosystems are impacted by various threats such as oil spills, algal blooms, and sediment floods, which disrupt habitats, wildlife, and human activities. Advances in satellite imagery and Artificial Intelligence (AI) have enhanced our capabilities for early detection and mitigation of such hazards. In this paper, we propose a marine event detection pipeline for Earth observation satellites equipped with multi- or hyperspectral sensors. Our approach includes a self-supervised neural network encoder that compresses satellite images into a reduced latent space, enabling efficient onboard processing. A machine learning anomaly detection model identifies deviations from normal sea patterns to detect environmental anomalies. We compare its performance against traditional algorithms such as Isolation Forest, One-Class Support Vector Machine and Local Outlier Factors. Our lightweight, resource-efficient pipeline is optimized for deployment on satellites with limited computational resources, ranging from embedded CPUs to AI hardware accelerators. By prioritizing the transmission of critical information, our solution enhances system responsiveness and optimizes satellite communication bandwidth. Demonstrated through current integration across multiple missions, including European Space Agency's (ESA) Phisat-2 mission and Microsoft/Thales Alenia Space IMAGIN-e mission, our pipeline aims to improve marine environmental monitoring by providing timely alerts and efficient data reduction.

Many machine learning methods aim to approximate the lower-dimensional manifold on which the data lives. A desirable feature of such methods is that they should capture the epistemic uncertainty of this learned manifold. One model that achieves this is the Gaussian Process Latent Variable Model, in which a Gaussian Process (GP) mapping from the latent space provides an estimate of the uncertainty of the manifold. However, the effectiveness of this uncertainty estimation is limited by the mean-field variational approximation between the GP inducing points and the latent variables. In this work, we apply Amortized Structured Stochastic Variational Inference to allow the variational posterior for the latent space to be conditionally dependent on the value of the inducing points. We demonstrate that this more flexible variational posterior improves several metrics relating to the reconstruction of points on the data manifold.

Stationarity rewards memory, but after a change the same history can mislead. We ask when forgetting should be permitted. E-process-authorized Thompson sampling (e-ATS) gives each arm full-history and discounted Beta states. An anytime-valid e-process first authorizes the discounted state, then a reversible relevance score controls its influence. Before authorization, e-ATS exactly follows optimistic Thompson sampling (OTS). Under a Beta-Bernoulli prior-predictive stationary model, e-ATS's probability of ever departing from OTS is at most the chosen $α_E$, without fitted thresholds. Relative to e-ATS, removing authorization increased mean normalized dynamic pseudo-regret by $38.4\%$ on the registered suite but reduced it by $7.5\%$ on the literature-derived replay suite. Therefore, evidence controls when adaptation begins, not whether it always helps.

Success conditioning is a strategy for improving decision-making policies in stochastic environments; it updates a policy by increasing the probability of taking actions that yield successful outcomes. Success conditioning is common to many reinforcement learning applications, yet its limiting behavior and convergence rates are not well understood. In this work, we demonstrate that success conditioning converges to an optimal policy on a broad class of Markov decision processes (MDPs). We also derive convergence rates in some common settings. For discounted MDPs, we prove convergence within $\mathcal{O}(1/\varepsilon^p)$ iterations to an $\varepsilon$-optimal policy, where the exponent $p$ depends on problem data. For single-period MDPs, such a policy is obtained within $\mathcal{O}(\log(1/\varepsilon))$ iterations.

Masked discrete diffusion models offer a promising alternative to autoregressive generation, but iterative sampling can be costly, and intractable sequence likelihoods complicate reward fine-tuning. We introduce IDRF, a framework for reward fine-tuning of few-step masked discrete diffusion generators. Starting from a standard reverse-KL-regularized objective, IDRF replaces the intractable sequence-level KL penalty with inverse-distillation regularization. With an optimal auxiliary denoiser, we prove that the population inverse-distillation loss upper-bounds the sequence-level KL divergence to the reference distribution. IDRF optimizes a trajectory-based surrogate of this loss without reference-model rollouts, so the student keeps its own few-step sampler. We view few-step generation as a finite-horizon Markov decision process and optimize reward with a clipped policy-gradient objective over the student's trajectories. Across DNA, image, and text generation, IDRF achieves high reward with up to $32\times$ fewer denoising steps than the reference while mitigating reward hacking and preserving sample quality.

Neural network loss landscapes have many symmetries, which are preserved by gradient flow but broken by finite-stepsize stochastic gradient descent (SGD). A canonical example of such a symmetry is scale in homogeneous networks: one can scale up the parameters in one layer and down in the next without changing the network output. Previous work has documented cases in which SGD breaks this symmetry in favor of balancing gradient noise or minimizing fluctuations. Here, we show that the solution geometry of undercomplete linear autoencoders instead selects a preferred sign for scale drift: on the PCA solution manifold, SGD favors large decoder weights. This directed scale drift occurs on a slow timescale, and its dynamics admit an analytically-tractable effective description. However, it cannot continue indefinitely: increasing scale eventually drives the dynamics towards a finite-stepsize stability boundary. The resulting solutions are sharper than a balanced baseline in the sense of the maximum eigenvalue of the loss Hessian, but different sharpness measures can move in opposing directions. Thus, undercomplete autoencoders give a concrete illustration of how loss geometry can convert residual gradient noise into directed motion along a manifold of functionally-equivalent solutions.

Large language models (LLMs) are increasingly used to support legal practice, education, and research, yet their reliability in national legal systems outside the United States remains largely undocumented. We introduce an expert-validated benchmark for evaluating LLM reliability on the Colombian legal system. The benchmark comprises 1,042 items spanning ten areas of law and three question formats (closed multiple-choice, semi-open, and open-ended IRAC), built through a human-in-the-loop pipeline with multi-stage expert review. We evaluate 15 contemporary proprietary and open-weight models with format-appropriate metrics. Accuracy on closed questions ranges widely, from 0.905 (Gemini 3.1 Pro) to 0.577, but on free-text legal answers factual correctness never exceeds 0.45 (on a 0-1 scale) for any model. We find a dissociation between answer relevancy and correctness (Spearman rho = -0.46): models reliably sound responsive while frequently being wrong, a pattern of particular concern for non-expert users. Closed-question accuracy and free-text correctness are strongly rank-correlated (rho = 0.94), so cheap multiple-choice screening predicts model ranking but overstates absolute reliability. An independent rubric-based LLM judge and blind human expert scoring both reproduce the free-text ranking (rho >= 0.88). The judge further reveals that only about half of the norms models cite are correct; the rest are wrong or non-existent. Reliability varies systematically by legal area and follows an inverted-U across question complexity. Our results indicate that current LLMs require expert supervision for Colombian legal tasks, and that grounding answers in authoritative sources is a promising path to higher reliability. We release the benchmark construction pipeline to support reproducible evaluation.

Camera-controlled video models are rapidly advancing toward long generation horizons and complex camera control. A key failure mode is 3D inconsistency: as the camera moves, objects lose permanence and scene structures shift. Existing post-training techniques, which assign a single scalar reward to the entire generation, are poorly suited to correcting these inconsistencies over long horizons. We introduce LoGo, which blends global and spatially localized rewards for camera-controlled video models. The local reward provides fine-grained credit assignment, which substantially improves 3D consistency, while the global reward preserves camera following and video quality. Across three base models, LoGo shows a clear advantage on DL3DV and TrajectoryBench, a new benchmark for long-horizon, complex-camera-control generation that current evaluations lack. LoGo effectively reduces local object shifts, artifacts, and global scene changes, illustrating the importance of credit assignment in post-training video models. Project website: https://ziqi-ma.github.io/logo-website/

Terminal-using agents benefit from reinforcement learning (RL) in coding, debugging, and other multi-step terminal tasks. In these tasks, later commands often depend on information or intermediate results produced by earlier commands. However, existing trajectory-level and step-level credit assignment methods do not explicitly trace the read-write dependencies through which commands affect the final outcome. Consequently, training signals could still be assigned to irrelevant operations, weakening learning from relevant steps. In this paper, we propose Dependency-Aware Group Policy Optimization (DepGPO), which uses execution dependencies between commands to guide credit assignment for terminal agents. Specifically, we construct a command dependency graph from execution traces and trace backward from the resources inspected by the task verifier. We then assign credit to relevant writes and their supporting reads along these paths, and use it to redistribute trajectory advantages across steps. Extensive comparative experiments and ablation studies demonstrate that DepGPO improves task performance and training stability on complex terminal tasks.

Physical fidelity has received increasing attention in world models and video generation, yet how video representations encode physical information remains less understood. We introduce the World Embedding Benchmark, comprising 8,000 controlled simulation cases from 80 families spanning fluid mechanics, solid mechanics, dynamics, and optics & electromagnetism. Each case pairs a rendered video with simulation-derived physical annotations, supporting three complementary tasks: text-video retrieval, physical-property regression, and multiple-choice video-description pair classification. We use these tasks to distinguish cross-modal physical alignment from the recoverability of quantitative physical information. Evaluated pre-trained omnimodal embedding models show weak retrieval and near-chance within-family pair classification, while lightweight probes recover useful physical information from frozen video embeddings. Continual contrastive training with physics-specific video-text pairs improves retrieval and pair classification but degrades physical-property regression, revealing a trade-off between alignment and quantitative information recoverability. Finally, we use the embeddings to retrieve reference videos for retrieval-augmented generation with MiniMax-H3. Retrieved references improve the physical fidelity of generated videos, with stronger retrieval models yielding larger gains in our experiments. Together, these findings highlight the need to evaluate physical alignment and property recoverability jointly, and demonstrate the utility of physical representations for improving video generation.

Designing a scientific instrument tests whether language-model agents can do physics rather than recall it, provided the grading cannot be argued with. We introduce NeutronGym, to our knowledge the first executable environment for neutron instrument design: agents build instruments through validating tools, McStas ray-traces what they build, and a level-resolved ladder grades syntax, runtime, structure and science with no LLM judge. Procedural families supply unlimited instances of a fixed layout whose design parameters the agent must set, with held-out parameter regimes; a curated slice, McStasBench, adds 16 tasks from published instruments behind memorization probes and a sandbox. Seven models reproduce at most 7 of the 16, none retrieves a reference, and none meets an improvement target. The environment also trains. Reinforcement learning on its reward takes Qwen3-8B from 11% to 77% of held-out instances of a family whose targets come from a hidden design (69% at a second seed), past an untrained Qwen3-32B, and the recipe holds, at one seed each, on three further gated families. The analysis says what that gain is. Without the ladder's partial credit it collapses by 60 points. From reward alone the trained model reaches what a classical optimizer reaches, at the agent's simulation budget, only when handed the closed-form physics (77% against 81%, a gap that does not separate at this size), while frontier models still solve 98-99%. Getting a trustworthy result meant failing four task designs that no-model baselines could solve, and we release the probes that found them.

The recent wave of one-step generative models, which compress the multi-step trajectory of diffusion via either distillation or learned flow maps, has reached an inflection point where they can generate high-quality images. Here, we ask a natural question that follows from these advances: what happens to the denoising trajectory of multi-step diffusion when generation is compressed into a single forward pass? We offer an empirical observation we call \textit{depth as time}: the denoising computation that multi-step diffusion performs across sampling steps appears to unfold across the depth of a single forward pass, and can be recovered by decoding intermediate layers with the model's own output head. Most interestingly, we show that this depthwise computation depends on the transport task a flow map is trained to solve. The most surprising case is MeanFlow, where probing shorter transport intervals reveals both denoising and renoising within a single network evaluation. In contrast, generators trained without a time-indexed transport task, such as drifting models, do not exhibit the same depthwise denoising. Consequently, we show that models that exhibit the depthwise denoising phenomenon are more compressible across the layerwise computation: a MeanFlow \texttt{SiT-L/2} model can be compressed by $16.6\times$ in parameters into a single time-conditioned block. We offer an explanation for this denoise-then-renoise behavior and show that, when we treat the layerwise computation explicitly as a flow, a single time-conditioned block can be trained to denoise across layers, compressing a MeanFlow \texttt{SiT-L/2} model by $16.6\times$ in parameters. Together, these results suggest that the temporal computation of diffusion is not eliminated by one-step generation, but reorganized across network depth.

Relational databases are among the most widely deployed forms of structured knowledge, and natural language access to them requires grounding language onto schema entities and relations while handling the ambiguity inherent in how people phrase requests. Existing synthetic NL-to-SQL data generation methods largely ignore this ambiguity and produce oversimplified queries that fail to prepare models for the complexity of real-world structured knowledge access. We present FALCON, a framework that generates realistic, ambiguity-aware NL-to-SQL data matching the complexity of challenging real-world benchmarks, at low cost using compact open models. Our approach combines reserved-word SQL seeding and persona-based prompting to generate structurally complex queries, while alignment-based filtering preserves difficulty by distinguishing genuinely incorrect examples from complex but valid queries. Human evaluation confirms consistent high quality across model sizes, and our generated data exceeds existing benchmarks in both SQL complexity and natural language richness. Difficulty-stratified analysis shows models trained on FALCON data increasingly outperform baseline-trained models as query complexity increases, validating our pipeline's success in generating challenging training data. When combined with a small proportion of existing benchmark data, mixed training recovers performance on simpler queries while preserving these advantages on complex ones. The model- and database-agnostic design enables organizations to generate high-complexity NL-to-SQL training data locally without external APIs.

There has been a surge of recent work on correlated equilibrium concepts in Markov games. However, existing results focus on concepts weaker than normal-form correlated equilibria (NFCEs), leaving open the more challenging question of computing such equilibria, which goes back to the seminal work of Papadimitriou and Roughgarden (JACM'08). Here, we establish the first efficient algorithm for NFCEs in finite-horizon Markov games with a fixed number of players $n$. In particular, with $S$ states, horizon $H$, and at most $A$ actions per player, it computes an $ε$-NFCE in time $S(AH/ε)^{O(n)}$. This is the first algorithm polynomial in $1/ε$ and the description of the game for NFCEs in an interesting class of problems beyond the normal-form setting. Moreover, under the usual assumption that recommendations are independent across states, we show PPAD-completeness---that is, computational equivalence to Nash equilibria---either in many-player games or when the precision is exponentially small. The key idea behind our approach is to run backward induction on a sequence of auxiliary stage games, but with the twist that in each step we compute a constant-expectation correlated equilibrium. This is a natural refinement of correlated equilibrium in which the conditional expected payoff from obeying is independent of the recommendation. In fact, our reduction goes both ways, establishing an equivalence between constant-expectation CEs and NFCEs in Markov games. For a fixed number of players, we observe that a constant-expectation CE can be computed approximately by combining linear programming with suitable discretization. In contrast, it is PPAD-hard in i) polymatrix (many-player) games at constant precision, and ii) two-player games at exponentially small precision. The latter result follows from an unexpected connection to rank-2 two-player games.

Online reinforcement learning (RL) enables robot policies to improve through physical interaction, but the assistance they require changes as their competence evolves. Existing intervention strategies based on offline estimates or fixed decision rules can therefore become mismatched to the current policy. To address this, we propose UniIntervene++, an adaptive intervention agent that learns to allocate control between autonomous execution and heterogeneous assisted behaviors during online RL. Specifically, UniIntervene++ first formulates the evolving RL policy, trajectory correction, and a task-structured CodePolicy as Options in a unified semi-Markov decision process and learns their relative values online. Building on this, competence-adaptive intervention periodically probes the RL policy through unassisted execution, keeping control allocation responsive to its evolving capability. Finally, coupled experience learning allows assisted behaviors to improve the RL policy, whose evolving outcomes in turn reshape future intervention decisions. In this way, UniIntervene++ jointly determines when to intervene, how to intervene, and when to return control as the RL policy improves. Across five real-world manipulation tasks, UniIntervene++ achieves an average success rate of 89.67%, outperforming all baselines by at least 6 percentage points, while reducing human intervention to 0.77%, a relative reduction of at least 94.6% from the best baseline. Code is available in our \href{https://github.com/dannyyudong/An-Adaptive-Intervention-Agent-for-Efficient-Real-World-Reinforcement-Learning}{GitHub repository}.

Continuous emotion regression estimates moment-to-moment valence and arousal while a viewer watches a video. In familiar-video deployment, responses fron training participant-specific estimate, and prior-dominating fixed fusion tests whether physiology adds residual correction. In five-fold subject-held-out evaluation on 24was within 0.05 and 0.32 MAE of fusion in the internal and external evaluations, respectively. Source-explicit ablations showed that video identity and within-video tine accounted for most of the reduction, while EG-FNIRS gains were smaller and varied across participants and videos. These results identify the video-time prior as a strong, low-cost baseline and position EEG-fNIRS as an optional residual signal for familiar-video emotion regression.

Temporal abstractions, often instantiated as options, have long been regarded as a mechanism for accelerating credit assignment, facilitating exploration, and enabling generalisation in reinforcement learning (RL). However, developing general option discovery methods that are effective in large-scale, high-dimensional domains remains a fundamental challenge. Existing option discovery methods are either confined to relatively simple domains, depend on handcrafted or quasi-symbolic representations, or offer little improvement over learning without options. We present Wayfarer, a general, domain-agnostic, online deep RL agent that discovers options through Laplacian representation learning from high-dimensional observations and leverages them for control. We show that the resulting options simultaneously improve exploration, accelerate credit assignment, and generalise effectively to unseen settings, enabling substantially faster learning of complex policies. Wayfarer achieves state-of-the-art performance among single-stream agents on the most challenging Atari 2600 games, with the largest gains in games that require long-horizon exploration and strategic behaviour, such as Montezuma's Revenge and Private Eye.

Reinforcement learning (RL) is increasingly used for financial optimal-control problems when complex dynamics make analytical strategies difficult to obtain. There are financial mathematics literactures which provides many solved models whose equations and controls could evaluate and guide learning; we ask whether RL can exploit these results. We place a proximal policy optimisation (PPO) agent in an analytically solved continuous-time broker--trader game. PPO replaces the broker and chooses its trading speed while interacting with an informed trader and stochastic uninformed order flow. We derive a finite-step reward from the broker's continuous-time payoff and verify its discrete implementation through grid refinement and an exact one-step identity. With zero uninformed flow, a validation-selected PPO--FFNN approaches the reference action. With stochastic uninformed flow, the tested PPO--FFNN and PPO--LSTM remain inaccurate, although supervised learning confirms that their actors can represent the action. Monte Carlo diagnostics show that their critics do not reliably rank nearby actions; potential-based reward shaping also gives no reliable improvement. Under partial information, a causal certainty-equivalent controller based on the broker's observable history remains close to the reference, while PPO has larger errors and lower payoffs. Finally, we freeze the analytical policy and train PPO to adjust it after the execution cost changes. Halving the cost yields a repeatable improvement that closes \(2.22\%\) of the gap to the changed-cost reference. The analytical solution therefore provides both a benchmark for diagnosing RL and a useful starting policy for adaptation.

Federated learning lets many clients train a shared model together without ever sending their private data to a central server. Each client shares only a model update, and this update should reveal far less about the client than its raw training examples would. This premise is what protects the privacy of the clients. Gradient inversion attacks challenge it directly by trying to reconstruct a client's private input images from the single update it shared. Under FedAvg, a client's update accumulates several local training steps, so the server sees only the two endpoints of a hidden weight trajectory. Recent gradient inversion attacks fit a surrogate model along the path between these two endpoints but they still read its gradient at a single point. We propose the Path-Integral Surrogate Model Extension (PI-SME) which treats the accumulated update as a path integral of the gradient field and approximates it by Gauss--Legendre quadrature over several nodes along a learnable Bézier path. On CIFAR-100 and FEMNIST images across a range of trajectory lengths and class-restricted batches PI-SME reconstructs the private inputs more faithfully than the strongest surrogate baseline on several inversion metrics and the matching loss.

Understanding and assessing natural hazards is essential for disaster preparedness and risk reduction. Recent advances in large language models have spurred growing interest in AI agents for hazard analysis, particularly their ability to integrate scientific data, models, and tools into automated workflows. However, effective automation requires agents to determine which scientific methods are appropriate for a given event and executable with the available data and tools. As new evidence and execution results become available, these conditions can change, requiring agents to reconsider their choices. We formulate this problem as state-dependent scientific route selection and introduce HazardWeaver. Specifically, HazardWeaver first leverages the Hazard Knowledge Compiler to extract evidence-linked conditions governing scientific applicability, then its Hazard Capability Graph represents executable scientific capabilities and checks compatibility between their inputs and outputs. Using these complementary representations, the Hazard Weaver Agent component selects applicable and executable routes, carries out their workflows, and revises its decisions as the analysis state changes. To evaluate both the scientific outputs and the decisions that produce them, we introduce the Hazard Weaver Benchmark, comprising 141 instances across seven single-hazard domains and four multi-hazard interaction classes. The benchmark accommodates multiple valid scientific routes and evaluates output correctness, route validity, and justified abstention. Extensive experiments on this benchmark show that HazardWeaver outperforms existing agent systems, with the largest gains on tasks with multiple eligible scientific routes. Our code is publicly available at https://github.com/LabRAI/HazardWeaver.

Security benchmarks for LLM-based agents often report the attack success rate (ASR) as a measure of model robustness and use these scores to compare different models and defense mechanisms, assuming that they describe the security of the agent. In this paper, we explore whether it also influences the benchmark's measurement. To measure the effect of the benchmark representation, we introduce threat-preserving representation sensitivity (TPRS), which measures how much the ASR changes when we change the agent-visible representation while holding the underlying task, harmful action, security policy, ground truth, environment, and the evaluation criteria fixed. On Agent Security Bench (ASB), replacing threat-related tool names with threat-neutral names raises the committed attack success rate by 11.67 percentage points on GPT-5-mini and by 13.21 points on Claude Haiku 4.5. On MCPTox, replacing the original neutral tool name with an explicit threat-related name lowers the ASR by 11.00 percentage points on GPT-5-mini and 4.11 points on Claude Haiku 4.5. On AgentDojo, adding threat-related wording to the attack-relevant tool changes ASR by only 0.50 percentage points on GPT-4o-mini, yet the benign utility falls by 5.36 points on tasks requiring that tool. We ran an experiment on MCPTox where we observed that a threat-neutral name matched on token count, length, and casing reproduces most of the shift produced by the threat-explicit name (8.54 of 11.00 points on GPT-5-mini). The results show that a security score measured under one representation may fail to generalize across threat-preserving representations of the same security problem. Robustness claims should therefore be supported by performance across a controlled set of threat-preserving representations rather than relying on a single representation-dependent score.

Diffusion Transformers (DiTs) can generate high-quality images and videos, but generating each sample requires multiple costly DiT forward passes. Two common ways to accelerate DiT sampling are step distillation, which reduces the number of sampling steps, and caching, which skips some DiT evaluations by reusing a tensor computed at an earlier step. Most caching methods decide in advance which tensor to reuse. After distillation, adjacent sampling steps are farther apart. Reusing a tensor across this larger gap introduces more error, so choosing what to cache becomes especially important. We therefore introduce AutoTarget, a method that chooses the cached tensor for a given model, solver, and reuse schedule. AutoTarget uses a small set of runs without cache reuse to measure the error caused by reusing each candidate tensor, then selects the candidate with the lowest error. We also analyze how an error at one reuse step affects the final sample. For Euler sampling, we identify cache targets that produce the same trajectory and show why a stored solver update may not. Experiments on distilled image and video DiTs show that the best cache target changes with the model, image resolution, and solver. AutoTarget reduces DiT evaluations and retained cache storage. Generation quality remains close to the corresponding uncached run. On the tested PixArt-LCM and FLUX.1-schnell settings, its calibration ranking matches the ranking from held-out cached runs. To help others reproduce the method, we provide its core implementation on GitHub at https://github.com/wali1024-offical/AutoTarget.

We introduce HyperBrowseComp, a multilingual and multimodal browsing benchmark comprising 423 manually authored and human-validated questions across 13 languages, written by native or highly proficient speakers. Questions are designed to be extremely challenging. Each question targets a concise, publicly verifiable answer whose discovery requires locating obscure evidence, following multi-step clue chains, or inspecting heterogeneous sources such as videos, scanned documents, images, or maps. Easier questions are filtered out by evaluating them with models without internet access to reduce the likelihood that they can be answered with parametric knowledge alone. We evaluate several models using provider-native search and a shared external retrieval harness under a common agent protocol. To contextualize model performance and effort, we also conduct a human evaluation on a sample of the questions. HyperBrowseComp provides a challenging testbed for persistent information seeking across languages and evidence modalities, with difficulty arising from discovering and connecting evidence on the open web.

Contactless heart-rate sensing with millimeter-wave (mmWave) radar requires assessing whether individual measurements support reliable estimation. We study learning to assess heartbeat observability, defined as the readability of the heartbeat component in an acquired phase spectrum, for selective heart-rate estimation. Coherent superposition of scatterer returns can suppress this component even under similar macroscopic observation geometry, motivating assessment directly from acquired measurements. To obtain training supervision across different observability conditions, we develop a controllable multi-scatterer frequency-modulated continuous-wave (FMCW) simulator. Agreement between the dominant heartbeat-band peak and the known heart rate provides an automatic observability label for each simulated measurement. We propose HEAR (Heartbeat Estimation with Assessed Reliability), a compact dual-task Transformer that jointly predicts an observability score and heart rate. Its input combines spectral magnitudes with frequencies relative to the respiration fundamental, providing context for respiratory harmonics. Trained solely on simulated observations, HEAR transfers zero-shot to two public real-world datasets collected at 60 and 120 GHz from 134 subjects. The same learned score supports selective prediction with both HEAR's own heart-rate head and multiple existing estimators. On the 120 GHz dataset, score-based selection reduces the HR head's mean absolute error from 17.9 BPM at full coverage to 1.6 BPM at 50% coverage. The complete pipeline achieves an end-to-end processing latency of 50.8 ms on an edge device. Project page: https://yuxuanhu9.github.io/HEAR/.

We describe the Writerslogic team's participation in the CLEF 2026 SimpleText shared task, addressing Task 1 (text simplification) and Task 2 (complexity spotting). For Task 1, we develop a multi-candidate generation pipeline using GPT-4o-mini that produces five simplification candidates per sentence at varying temperatures, then selects the best candidate using a reference-free scoring heuristic that rewards compression, source word retention, Cochrane Plain Language Summary vocabulary usage, and lexical simplicity. On Task 1.1 (sentence-level simplification), our Claude Sonnet 4 submission achieves SARI 47.43 and BLEU 14.21, the top-ranked sentence-level system (3rd on the combined Task 1 leaderboard, behind two document-level submissions). For Task 2, we fine-tune a DeBERTa-v3-large NLI model on 350K labeled (source, sentence) pairs, framing hallucination detection as natural language inference. The model reads the most relevant source sentence as premise and the candidate as hypothesis, directly learning to distinguish grounded from hallucinated content. On Task 2.1 (binary overgeneration identification), our fine-tuned DeBERTa system achieves 0.8081 document-level macro F1 (0.8085 in our best ensemble), the top-ranked entry within the identification track and 2nd among teams overall, behind AIIR Lab (0.8197). On Task 2.2 (multi-class error classification), our best submission reaches 0.804 multiclass accuracy, ranking 2nd among unique teams behind AIIR Lab (0.827). We evaluate both tasks on English and multilingual biomedical text from Cochrane systematic reviews.

We describe the Writerslogic systems for three PAN at CLEF 2026 shared tasks (Reasoning Trajectory Detection, Voight-Kampff Generative AI Detection, and Multi-Author Writing Style Analysis), unified by a shared analytical framework: feature robustness under distribution shift is governed by support overlap between training and test distributions, not by training-set effect size. This yields a taxonomy (domain-anchored, domain-portable, domain-invariant) that explains why generator-specific features die under domain shift while vocabulary fingerprints (hapax ratio, Yule's K, Heaps' exponent), compression measures, and character n-grams survive. On Reasoning Trajectory Detection, where training was entirely mathematics and 84 percent of test was unseen domains, the framework guided system design to 1st place in source detection (0.85 macro F1 via Opus-Sonnet agreement) and 3rd place in safety classification (0.66 macro F1 via query-refusal decomposition). For Voight-Kampff, we built a calibrated ensemble of DeBERTa-v2 (ONNX), multi-seed LightGBM with 44 domain-portable stylometric features, and SVM on n-gram TF-IDF, combined via learned stacking with isotonic calibration; the best configuration achieved 0.891 on the PAN 2026 test set with balanced sub-metrics (0.853 to 0.902 across all evaluation dimensions). For Multi-Author Writing Style Analysis, we describe a system fusing spectral clustering over character n-gram similarity graphs, normalized compression distance for local boundary detection, and SmolLM-135M perplexity for neural change-point detection; a platform mix-up meant our run never reached the official evaluation, so we report the design and its a priori predictions. Across all three tasks, features measuring generation process properties are designed to outperform features measuring generated content properties under domain shift.

Financial AI agents must do more than retrieve facts: investment workflows require correct quantitative execution, reliable use of procedural resources, and auditable structured outputs. We introduce FinSkillBench, an evaluation suite of 2,603 point in time episodes across 12 subtasks in portfolio construction, risk management, and fundamental analysis, with hidden regenerable ground truth and task specific deterministic verifiers. Executing 17,820 episodes across 9 models and 3 resource conditions, the paired analysis across 8 models shows that curated skill packages raise mean scores by +16.2 points (0.366 to 0.528), whereas skills generated within a single episode add only +0.5 points while consuming more tokens and turns. We then decompose the curated premium by granting human authored procedural documents and executable domain tools separately: documents alone add +5.6 points, tools alone add +19.5 points, and their combination is subadditive. The premium is strongly workflow dependent: executable tools dominate numerically intensive workflows, documentation matters more when procedural or output schema guidance is the bottleneck, and interpretive tasks benefit from both. The effects are sign stable across 10 scoring variants and cluster bootstrap analyses, and an independently implemented second harness reproduces the directional pattern while showing that effect magnitudes depend on how tools and data are exposed. Overall, a measured "skill premium" is a property of the full model, resource, and harness system rather than of the underlying model alone.

Cephalonauts One is a whole-brain 3 Tesla (3T) functional magnetic resonance imaging (fMRI) dataset recorded while subjects listened to audio podcasts. Three healthy subjects underwent multiple scanning sessions, each consisting of five 15-minute runs, while listening to podcasts in their native language. With 30 hours of fMRI data per subject, the current release is the deepest available fMRI dataset using naturalistic speech stimuli. The dataset pairs brain activity with the corresponding podcast audio, transcript annotations, and derived stimulus embeddings. Furthermore, we introduce a brain decoding benchmark formulated as audio segment retrieval: given fMRI activity from a held-out session, the decoder must identify the corresponding time-aligned podcast audio segment among candidate segments. We provide standardized splits, evaluation metrics, and baseline decoders for this task. Finally, a scaling analysis shows that decoding performance improves continuously with the amount of training data per subject.

Empirical claims about the connection between over-squashing and long-range interactions in GNNs, can only be trusted if the benchmarks used to validate them genuinely require long-range interactions. The de-facto standard, the Long Range Graph Benchmark, has been repeatedly shown to be saturated by tuned short-range models, with existing synthetic alternatives being tied to specific topologies. As such, there is a lack of principled certificate of long-rangedness on arbitrary graphs. This state reflects the absence of a precise characterization of long-ranged benchmarks. We address this fundamental gap by introducing four verifiable axioms: Predictability, Tightness, Strictly $k$-Range, and Topology-Invariance, that any task claiming to test $k$-hop interactions must satisfy. We formally prove that violating any one of them admits failure modes that undermine conclusions drawn from the task. Based on these axioms, we introduce TRIP (Truly Ranged Interactions Problem) and its generalisation GRIP (Generally Ranged Interactions Problem), constructive procedures that turn any graph into a provably long-ranged task by drawing features from stable distributions. Moreover, by construction, GRIP admits a closed-form, per-range Maximum-Likelihood oracle that yields the first a priori per-range lower bound on test error available on any benchmark. Using our framework, we: (i) audit 4 common long-range benchmarks and identify their failures modes with respect to our axioms; (ii) on TRIP-instantiated topologies, we find a popular notion of curvature is uncorrelated with GNN performance, supporting topological-vs-computational bottleneck distinction; and (iii) we show that a novel benchmark's over-squashing measures factors beyond pure long-rangedness. Code to use the framework and reproduce experiments is released https://github.com/ferranhernandezc/graph-grip.

Large language models (LLMs) have reached expert-level performance on competition mathematics largely through the volume of search placed around them: candidate solutions are sampled in quantity and retained only when an external criterion accepts them. Such a procedure improves the outcome that survives it while leaving untouched what the model represents. We examine that question where no external criterion exists: translating statements between the dialects of neighbouring subfields, where fidelity turns on the level of generality at which content is asserted. The source leaves that level implicit in its vocabulary, so a faithful translation must recover it from the relation between the theories. We introduce an instrument that codes truth, content and scope in separate blind queues, with a judge-free measure of whether a rewrite states the hypothesis implicit in its source, and establish its sensitivity with a planted-positive control. Across seven models from four families, translating towards the general framing widens the domain of quantification in 60.6% of rewrites and narrows it in none; translating towards the concrete framing narrows it in 28.3% and widens it in 0.3%. The hypothesis that would prevent it is stated in 21.6% of model rewrites and 4.2% of human statements. Capability does not govern the asymmetry: it appears in every model tested, and the most capable widens least. It replicates on the half of the benchmark held out by a pre-registered rule, and on statements written by mathematicians. Instructing a model to state every hypothesis it requires raises that rate but not its sensitivity to direction. We argue that these systems have acquired an object-level correspondence between subfield vocabularies without the constraint under which a translation between theories carries hypotheses to hypotheses.

Generating progressively harder reasoning problems requires synthesis procedures that adapt as the task distribution evolves. Existing task-level recursion reuses generated problems as seeds but leaves the construction harness unchanged. We present task-harness co-evolution, a framework for recursive harness self-improvement (RSI) in reasoning-data synthesis. Online self-improvement converts intermediate solver failures into reusable skills during generation. Post-task self-improvement revises skills, prompts, and workflows after each batch, adopting candidates only when they generate harder valid tasks within a bounded cost increase. Model weights and verification criteria remain fixed. Across mathematics, coding, and science, mean solver accuracy decreases from 100.0% to 54.8% over fourteen evolution rounds. Ablations show that combining both update schedules produces harder tasks than fixed-harness recursion or either schedule alone. The resulting data improves downstream SFT and GRPO performance. In particular, a 27B student fine-tuned on 10K synthesized mathematics examples achieves 62.5% mean-16 accuracy on APEX, competitive with selected frontier-model references. These results support adapting the synthesis harness alongside the tasks to generate increasingly challenging data with downstream training value.

EEG foundation models (EFMs) capture reusable knowledge from large-scale EEG data, while many EEG recordings also include companion physiological signals that provide complementary information beyond the EEG-only interface. The challenge is to preserve this pretrained knowledge while extending the EFM to heterogeneous multimodal recordings through an adaptation inferred from unlabeled target data. We introduce ZeroMAG, a zero-shot multimodal adapter generation framework that extends a frozen EEG encoder and prediction head using unlabeled target recordings, without target labels or target-side optimization. The target datasets are held out from all model training and selection in the ZeroMAG pipeline. ZeroMAG organizes companion modalities around a configuration-invariant adapter, constructs a modality-subject-task condition from unlabeled recordings and task context, and generates adapter weights in a function-constrained latent space learned from source adapters. Across six held-out target datasets and three EFM backbones, ZeroMAG improves balanced accuracy by 7.22 percentage points over EEG-only inference and 4.89 points over direct weight regression, while coming within 0.50 points of supervised multimodal adaptation on average. Ablations further show that removing functional supervision from either representation learning or conditional generation degrades generated-adapter performance, confirming the contribution of both components.

Endovascular brain-computer interfaces (BCIs) avoid craniotomy but require precise device delivery through anatomically variable cerebral veins. This work presents the first demonstration of in vitro autonomous robotic navigation for endovascular BCI access in the cerebral venous system. Soft Actor-Critic controllers were trained in silico for two sequential tasks spanning the right internal jugular vein to the superior sagittal sinus, using geometric augmentation of one training anatomy. Navigation was evaluated in a training anatomy and an anatomically unseen hold-out model over 250 in silico episodes and five fluoroscopy-guided in vitro robotic runs per task-anatomy condition, comprising 1,000 simulated episodes and 20 physical runs overall. Task recurrent predictors were also evaluated for online identification of impending navigation failure. In silico success rates for Tasks A and B were 85.6% and 98.4% in the training anatomy and 42.0% and 91.6% in the hold-out anatomy, respectively. Fourteen of 20 physical runs were successful (70% overall), including 80% success for Task B in the hold-out phantom. In silico the predictors detected 99.3-100.0% of failures with false-alarm rates of 0.8-6.7%. During in vitro evaluation, predicted risk increased before failed episodes, but elevated probabilities during some successful runs showed reduced calibration after transfer. These results demonstrate the feasibility of autonomous cerebral venous access and show how online failure prediction could support human oversight, while also identifying anatomical generalization and sim-to-real calibration as priorities before preclinical translation.

Authorship Attribution (AA) requires capturing fine-grained stylistic characteristics, making it particularly challenging in zero-shot (ZS) settings where no task-specific supervision is available. In this work, we investigate the effect of author representations on ZS AA by evaluating a label-only prompting baseline together with three author representation strategies: representative writing samples, LLM-generated descriptions, and style embeddings (LISA). The first three approaches perform attribution using LLM prompting, while the embedding-based approach uses style embeddings with cosine similarity. We investigate the influence of prompt design and propose a two-stage embedding-based attribution framework that combines candidate space reduction with embedding-dimension selection. The results show that label-only ZS AA is ineffective, while incorporating author-specific representations consistently improves attribution performance. Among the evaluated approaches, the proposed two-stage LISA framework achieves the strongest overall performance, whereas LLM-generated style descriptions provide a substantially more compact representation of author style at the cost of some attribution performance. These findings demonstrate the importance of author representation in ZS AA, while indicating that current open-source LLMs remain insufficient for robust attribution without more effective representation learning.

Distillation has become a core primitive of large language model training, but its properties are not yet well understood. We take an entropic perspective, studying how the entropy of the student depends on the data and the divergence that define the distillation objective. We prove that forward KL inflates the entropy of the student above that of the teacher. Since cross-entropy training is a special case, this yields an identity that we verify quantitatively in pretraining and supervised finetuning. Other divergences come with no such guarantee: reverse KL deflates entropy until the gap between student and teacher gets too large, and interpolating between the two changes entropy smoothly early in training but abruptly at convergence. The lower entropy of on-policy distillation comes from token-level reverse KL, not from on-policy sampling. The divergence therefore acts as an implicit entropy regularizer, whose role is clearest in self-distillation: as conditioning on privileged information deflates entropy, the divergence hyperparameters that work best are those that compensate for it.

Explainers for Graph Neural Networks (GNNs) are commonly evaluated by their plausibility, i.e., how well their explanations recover a predefined ground truth, such as a motif planted in the data. This protocol implicitly assumes that a GNN trained on such data relies on the intended motif. Although prior work has questioned this assumption, plausibility remains widespread. First, we show that the assumption is violated on several widely used benchmarks, where, e.g., degree statistics alone suffice to solve the task. Then, we remove this confounder by replacing training with compilation. We achieve this by introducing $\mathsf{Gracr}$, the first compiler translating graded modal logic formulas into GNN weights, yielding models that replicate the behaviour of the corresponding formulas. Since the behaviour of the model is now known by construction, we can define its ground truth explanation formally and compute it exactly. Building on this, we introduce $\mathsf{Gracr}\mathsf{Bench}$, a benchmark of compiled GNNs for the evaluation of explainers against this exact ground truth. Experiments on eleven explainers across six tasks show its effectiveness for fine-grained diagnostic evaluation: notably, we discover that most explainers are not robust to indirect influences or alternative implementations of the same formula. These results position $\mathsf{Gracr}\mathsf{Bench}$ as a novel, rigorous evaluation setting for graph post-hoc explainability.

Table-to-report generation refers to the task of automatically generating article-level analyt- ical reports from relational tables and is an essential capability for automated data science and decision support. Its central challenge lies in systematically discovering verifiable com- posite insights across tables, attributes, and analytical perspectives, and organizing them into coherent, complete, and traceable evidence chains. Existing methods primarily rely on sequential, reactive data agents or direct Large Language Model(LLM) generation. They suffer from exploration bias: early local observations constrain subsequent actions, causing models to focus prematurely on local analyzes and miss cross-table or cross-dimensional evidence. We propose ComInsight, which reformulates insight discovery as the composition of atomic evidences. We first define an atomic insight as the smallest executable analytical unit conforming to a predefined analysis pattern and enumerate all valid atomic insights from database schema and content. These atoms are then organized into a multi-relational insight graph, where nodes represent verified data facts and edges encode logical, temporal, or hierarchical relations. Finally, a set of composition operators systematically fuses atomic nodes into higher-order composite conclusions. Every composite output is accompanied by executable SQL and fine-grained provenance, ensuring full verifiability. Across three benchmarks InsightBench, DDR-Bench, and T2R-Bench, ComInsight consistently outperforms strong baselines in factual correctness, novelty, and structural completeness. We believe ComInsight offers a reliable, efficient, and explainable path toward table-to-report generation.

General-purpose Text-to-SQL systems achieve strong performance on academic benchmarks like Spider and BIRD, where schemas are relatively shallow and column values are often human readable. In production financial databases, where concepts are stored as opaque integer keys rather than human-readable strings, these methods fall below 50%, as even simple queries require multiple joins and filter predicates reference opaque IDs. We present Financial LINking Text-to-SQL (FLINT), a domain-specialized Text-to-SQL system that closes this gap through three key components: (1) a lookup agent that dynamically resolves natural-language concepts to question-specific reference table constraints, (2) embedding-based retrieval of structurally similar query templates from a compact, expert-authored bank, and (3) schema linking that prunes a large table schema to the relevant subset by traversing foreign-key chains, rather than relying on name similarity alone. We evaluate on two datasets totaling 359 questions over production financial schemas. FLINT outperforms various state-of-the-art baselines using the same LLM. The system is deployed in production as part of a financial data retrieval service.

A reasoning model asked whether a causal effect is recoverable from observational data can fail in two ways: it refuses an identifiable query or answers a nonidentifiable one. The latter is more consequential, as no observational data can validate the claimed formula. Measuring this failure requires queries that are provably non-identifiable, which prior evaluations lack, and grading that accepts correct formulas in any equivalent form, which string matching cannot provide. We build CERTID, a formal identification pipeline that addresses both limitations. CERTID uses the sound and complete causal identification algorithm ID to certify whether an effect is identifiable from a given graph and query, and verifies returned formulas against structural causal models whose interventional distributions are known exactly. CERTID further develops theoretical results to mitigate structural leakage, repair non-identifiable queries, and establish grading guarantees. We evaluate three frontier reasoning models (Gemini Flash, Gemini Pro, and GPT5.5) on 1,200 certified instances spanning 4 to 50 vertices. Accuracy proves a poor proxy for soundness: on identical instances, the false-claim rate on non-identifiable queries varies by seventeen-fold across models. We also find that models decide identifiability with 97-100% accuracy on graphs generated after the strongest model's training snapshot. Instances, the certification procedure, the verifier, and per-instance records are available at https://anonymous.4open.science/r/certid-D718.

Static-analysis scanners can identify personal-data types in source code, but they lack mechanisms to connect these findings to standardized privacy vocabularies. PrivDev maps 122 Bearer CLI data types to Data Privacy Vocabulary Personal Data (DPV-PD) categories and links them to potentially relevant GDPR provisions. Our approach combines deterministic mapping for 43 exact-label matches with a retrieval-grounded Large Language Model (LLM) to resolve the remaining 79 non-trivial mappings. The resulting RDF knowledge graph contains 118 ODRL policy resources that were structurally validated using SHACL. The artifact passed five complementary validation gates that cover structural correctness, query consistency, retrieval quality, LLM-based assessment, and human evaluation. In human evaluation, nine annotators produced 711 judgments, yielding a raw agreement of 0.72, Gwet's AC1 of 0.68, and Gwet's AC2 of 0.88. Our results indicate that the proposed mappings are plausible and reproducible, while also revealing ambiguities in scanner-defined data-type labels and coverage gaps in DPV-PD.

World action models (WAMs) have advanced robot control by predicting how observations and actions evolve over time. Despite this progress, RGB and action based future prediction does not explicitly address the spatial understanding needed for robot manipulation. Existing efforts often add a limited set of spatial prediction tasks through specialized heads or branches, leaving both the range of spatial supervision and the model architecture fragmented. We introduce XGenAct, a world action model that represents RGB observations, robot actions, metric depth, surface normals, and functional role segmentation as RGB videos through deterministic codecs. By sampling perception and action tasks during training, XGenAct uses one video diffusion transformer and one objective to learn temporal prediction across these spaces without modality specific learned heads. On held out RLBench tasks, structured perception training improves average closed loop success over RGB only training, and XGenAct achieves 52% success in the five task external comparison, versus 26% for the strongest evaluated baselines. It also predicts future depth and segmentation more accurately than the evaluated pipelines that generate RGB first and then apply a frozen perception expert.

On-policy self-distillation (OPSD) uses reference solutions as privileged hindsight to supervise student-generated reasoning trajectories. However, reference-based guidance may explain a correct solution without addressing why the student's own reasoning fails. This reasoning mismatch between the guidance provided and the correction needed can encourage the student to borrow correct conclusions while leaving its reasoning errors unresolved. Moreover, applying the same hindsight throughout the trajectory risks a distillation trap, where unnecessary constraints on valid reasoning compete with correction of substantive errors. To address these issues, we propose Root-Cause-Guided On-Policy Distillation (RC-OPD), which uses repairs of the student's own reasoning to provide guidance that addresses its specific errors while building on valid progress. For each failed attempt, RC-OPD locates the earliest substantive error, develops a local correction, and uses the corrected intermediate result as an anchor for the valid prefix. An iterative diagnosis--repair--continuation process tests the repairs through student continuation, identifying further errors within a fixed repair budget. For repair chains that reach a correct answer, root--cause--guided distillation uses failure diagnoses and corrective goals to supervise the erroneous segments, while anchor-guided distillation supports the corresponding valid prefixes with reasoning chains leading to the repaired intermediate results. We evaluate RC-OPD across multiple datasets and model scales. Extensive experiments and analyses show that it mitigates reasoning mismatch and the distillation trap, yielding substantial performance gains.

Recent advances in autoregressive video generation have improved temporal consistency over extended durations, yet interactive storytelling requires more than continuous scene extension: a new shot may combine characters and backgrounds from different historical shots. Whole prompt retrieval can overlook the distinct reference needs of individual components, while directly combining all historical memories may introduce unrelated visual content. To address these problems, we present Weave Forcing, a training-free framework for compositional memory reuse in interactive long video generation. First, we use an LLM for semantic slot routing to decompose user prompts into character and background descriptions and explicitly select suitable historical references for each component. To isolate the required content, masked memory weaving uses contrasting attention maps conditioned on semantic slots to construct refined semantic masks, selectively exposing relevant tokens from compressed historical KV memories to guide the generation of the current shot. We further introduce coverage adaptive RoPE to adjust temporal offsets and memory retention according to no, partial, or full reference coverage, addressing visual artifacts observed when incomplete historical references are positioned close to the current generation. Extensive experiments demonstrate that Weave Forcing improves cross-shot subject and background consistency while maintaining competitive visual quality and text alignment.

Chain-of-thought (CoT) reasoning allows humans to inspect how large language models reach their answers, and oversee model behaviour. This reasoning comes at an increased inference cost, motivating efficient methods that train models to solve tasks using fewer tokens. However, a common concern is that such training may cause models to skip important reasoning steps, so the CoT no longer faithfully reflects the model's decision. It is unclear whether or when this occurs in practice, since different efficiency methods apply length pressure to models' CoT in distinct ways, and faithfully explaining a model's decision takes more tokens on some tasks than others. To understand these dynamics, we fine-tune a variety of models with three methods that apply length pressure differently, namely a fixed generation budget, a per-example length target, and a group-relative length reward. We evaluate how efficient reasoning affects CoT faithfulness (i.e., how well the CoT reflects model decisions on related inputs) and monitorability (i.e., whether the CoT reveals when input interventions alter the output). We find that it affects faithfulness and monitorability differently. Faithfulness falls in most settings, primarily because the trained models are less consistent. Monitorability is more robust, as models keep acknowledging the influence on their answer even when the CoT is much shorter.

Post-exposure microscopy is central to qualification of fusion materials. However, manual analysis does not scale to the volume, heterogeneity, and multiresolution character of modern fusion-materials campaigns. To address this challenge, we present a reproducible workflow, implemented in the Galaxy scientific workflow environment, for automated crack identification and quantitative damage assessment from scanning electron microscopy images. The workflow processes SEM images and experimental metadata to identify cracks, quantify damage, and retain the intermediate products and processing history needed for reproducibility. Outputs include crack masks, skeletonized crack networks, quality-control visualizations, and scalar damage descriptors. The method is designed to operate without image-specific parameter tuning across tungsten grades, microstructures, magnifications, and damage states. We demonstrate the workflow on a sparse electron-beam thermal-shock dataset containing 418 images from 114 experiments spanning five tungsten grades and three microstructural states. We define a crack-density descriptor, which provides standardized inputs for downstream machine-learning prediction and physics-based crack simulation. These predictive components are exposed in the same Galaxy environment and are intentionally treated here as extensible workflow modules. The principal contribution is therefore an end-to-end, shareable, and computationally portable workflow that links experimental characterization, automated image analysis, preliminary damage prediction, and simulation-guided data acquisition for fusion-materials research.

Training-free posterior sampling methods, also known as Plug-and-Play methods, leverage pretrained unconditional diffusion or flow-matching models to solve inverse problems. Most existing approaches rely on guidance weights to balance, at each time step, prior information from the unconditional score or velocity network with measurement consistency, yet the tuning of these weights is often not discussed and is largely left to heuristics. We introduce a simple and principled offline strategy for automatically tuning these guidance weights. Our key observation is that, at each time step, the conditional denoising score-matching objective for diffusion models, or the conditional flow-matching objective for flow-matching models, is a least-squares objective. Therefore, when the conditional prediction is expressed as a weighted sum of the unconditional network output and a measurement-guidance term, optimizing over these weights reduces to a two-dimensional linear least-squares problem. The resulting time-dependent guidance weights can be optimized offline for a given measurement operator, noise level and sampler at the cost of a single minibatch of sampling trajectories, without retraining or fine-tuning the pretrained generative model. Instantiated with the standard Tweedie-based measurement-consistency term, our approach improves posterior sampling and achieves state-of-the-art reconstruction performance across diffusion- and flow-matching-based methods. Moreover, the optimized guidance weights enable diffusion samplers to reduce the number of sampling steps from 1000 to 50 with no significant degradation in reconstruction quality. Code will be made available.

Mechanistic edits (ablations, weight edits, activation steering) are the standard tools for unlearning a harmful capability from a neural network while preserving useful ones. Current approaches validate their effects only by testing, which can never cover an entire continuous region of inputs. Prior work at the interpretability-verification boundary certifies descriptions of a model: what a circuit computes, or whether it faithfully explains the whole. We instead certify the behavioral effect of an edit: that disabling a circuit removes one skill and provably preserves another, for every input in a region; a feature non-interference guarantee in the information-flow-security sense. We demonstrate such certified edits from toy ReLU networks up to a standard softmax + LayerNorm transformer, proving removal and preservation over continuous embedding-space regions and reaching roughly 9x the input-perturbation dimension an exact solver can handle by switching to sound bound propagation. Furthermore, we prove that no finite deterministic black-box test can certify removal, exhibiting an edit that passes exhaustive testing yet provably fails on a survivor pocket that can be made arbitrarily small. Guarantees hold on small, standard-architecture networks and, like any removal claim, presuppose that the target skill admits a decidable specification, a property which real-world harms may not have.

Deep tabular generative models are benchmarked on datasets with tens of thousands of rows; clinical datasets have hundreds. We preregistered and ran a size-ladder benchmark to find where the two regimes diverge: 8 public datasets subsampled from 200 to 20,000 training rows, seven generators (independent marginals, Gaussian copula, SMOTE, unconditional SMOTE, CTGAN, TVAE, TabDDPM) with a fixed 20-trial tuning budget and 5 evaluation seeds, plus 4 natively small clinical datasets at true size, for 2,220 committed runs in total. The primary metric is the AUROC of fixed classifiers trained on synthetic and tested on real data. In 23 of 24 (dataset, deep model) pairs no deep model ever beats the best trivial baseline by more than seed noise, at any training size we measured. The best baseline wins 40 of 49 (dataset, size) cells. Our preregistered prediction that the deep models' ranking would be unstable at small sizes is falsified: mean Kendall tau between adjacent rungs below 5,000 rows is 0.806, above our 0.8 threshold, and stability is highest at the smallest sizes rather than lowest. One caveat bounds all of this: in 81% of cells the gap between the top two methods is smaller than the variation between seeds. Finally, method rankings on natively small clinical datasets agree only moderately with rankings on subsampled large ones (mean tau 0.57 to 0.64), which questions whether a subsampled large dataset can stand in for a small one. All 2,220 result files, the preregistration and its hash, and the code that regenerates every figure and number from those files are public.

Adversarial patches can disrupt Vision-Language-Action (VLA) models by manipulating visual observations, leading to failures in robot control. However, it remains poorly understood which internal mechanisms underlie these failures and how targeted interventions can mitigate them. In this work, we mechanistically analyze VLA representations using a sparse autoencoder (SAE) and identify a feature whose activation strongly correlates with the presence of an adversarial patch. Based on this analysis, we suppress the identified feature at inference time only when a linear probe detects an attack. This intervention improves robustness without the cost of fine-tuning the VLA. We evaluate our method against VLA adversarial patch attacks on LIBERO-10. Conditional intervention improves success rate under intermittent attacks, whereas continuously applying the same intervention substantially degrades policy performance. These results show that attack-related internal representations can provide useful targets for VLA adversarial defense and that controlling when to intervene is important for limiting disruption to nominal policy behavior.

Long-term forecasting models commonly process all patches in a look-back window using the same fixed stack. Older contextual patches and recent evidence therefore receive the same computational depth. Yet the information closest to the forecast and the more distant context do not contribute equally. Uniform processing leaves this distinction unexpressed in the architecture. We propose MARO, a Most-Recent Anchoring with Recurrent Ordering model that processes the look-back window from the most recent patch to the oldest. The most recent patch serves as the anchor. It initializes the latent state and conditions each subsequent step, so older patches are folded into a representation that remains centered on recent evidence. A single shared module is reused at every step, so extending the scan further into the past introduces no additional parameters. Intermediate states retained during the scan allow the forecast head to weigh short and long portions of the history separately. This expresses recency through the order of recurrent refinement. Extensive experiments across multiple real-world time series datasets show that MARO achieves state-of-the-art performance on both long-term and short-term forecasting tasks.Ablation studies examine the contribution of the main architectural components.

Most long-term time-series forecasting models map the look-back window directly to the full horizon in a single pass. While efficient, this design does not explicitly identify which historical states are most relevant to different future segments or exploit what followed those states. Analog forecasting addresses this by retrieving past states similar to the present and using their observed continuations, but single nearest matches can be unreliable and overlapping patches may produce redundant candidates. We propose DuoTS, a Dual-Context Time Series forecasting model that uses retrieved evidence without relying on it exclusively. DuoTS first produces a base forecast with a parallel patch encoder and linear prediction head, then progressively refines it one future patch at a time. Each refinement combines two views: a current context that attends to recent tokens and captures the latest dynamics, and a detail context that provides distinct retrieved analogs together with their subsequent trajectories. Patch-wise refinement allows the model to balance these views across the forecast horizon and associate each future segment with evidence appropriate to its temporal distance from the present. Experiments on multiple real-world datasets show that DuoTS achieves state-of-the-art performance, while ablations confirm the contribution of each context. The refinement mechanism is also model-agnostic, requiring only an encoded look-back window and the future-patch position, and can therefore be integrated into existing forecasting models.

We introduce AREX, a training-free sampler for pretrained flow matching models that uses the target mean and covariance to capture an analytically tractable part of the sampling dynamics. We show that the velocity field of the moment-matched Gaussian target is the $L^2$-optimal affine approximation to the marginal velocity field. This motivates decomposition of the learned dynamics into an affine component over the whole sampling path, determined by the first two target moments, and a neural residual term. AREX keeps the affine component and integrates it using an explicit matrix-valued propagator. In turn, we only require to integrate over the residual term. This differs from scalar exponential integrators, which analytically handle only isotropic linear dynamics. Across image and text-to-image generation tasks, AREX consistently improves sample fidelity in the few-step sampling regime without retraining the underlying model.

Hallucination detection is particularly important for medical language models, but repeated-sampling approaches are expensive and existing uncertainty-head resources do not directly transfer to a new backbone and language. We adapt the LLM Uncertainty Head (LUH) framework to Aya-Expanse-8B-based Persian medical models, using Gaokerena-V and Gaokerena-R as two previously developed backbones. We first examine response variability on a 168-question Iranian medical entrance examination and observe substantially lower five-run consistency for Gaokerena-V than for Aya-Expanse-8B, whereas Gaokerena-R is comparable to Aya-Expanse-8B. We then construct two paired claim-level hallucination datasets directly in Persian, containing 1,600 responses for each backbone, and train lightweight claim-level heads on frozen backbone attention maps and token probabilities. On held-out test splits, the heads obtain PR-AUCs of 0.4820 and 0.4652, corresponding to 2.30 and 2.66 times their respective random baselines, and ROC-AUCs of 0.7852 and 0.7810. The heads require neither retrieval nor repeated sampling at inference time. These results provide an initial study of single-pass claim-level uncertainty estimation for Persian medical language models; the test splits are small and the labels are automatically generated.

Post-training a large language model (LLM) often requires exploring trade-offs between multiple rewards, but retraining for each trade-off is expensive. Decoding-time policy composition allows these trade-offs to be adjusted by combining reward-specific policies at inference time. This composition targets a weighted product of the policies' probabilities over complete responses, but standard implementations combine their next-token probabilities, generally introducing sampling bias. We analyze a known iterative correction based on independence Metropolis-Hastings (MH). Our main result shows that, for every rollout budget, MH produces an output distribution at least as close to the target as sampling-importance-resampling (SIR) with the same budget, as measured by every convex f-divergence. We also derive a lower bound on MH's improvement over the uncorrected decoder in a consensus objective measuring agreement with the supplied policies. We further characterize the correction's sampling error in two asymptotic regimes: when the reward-specific policies approach agreement, and when the log ratio between target and uncorrected-decoder probabilities fluctuates increasingly widely, as can happen for long responses. We complement our analysis with experiments in enumerable and LLM-scale settings.

Long-horizon mobile manipulation presents significant challenges due to compounding execution errors and capacity interference between locomotion and arm control. While recent Vision-Language-Action models excel at short-horizon tasks, they lack the hierarchical reasoning required for multi-stage objectives. Furthermore, existing hierarchical agents suffer from rigid sub-task mapping, inflexible replanning, and a lack of continuous learning. To address these limitations, we introduce MobiAgent, a dual-loop agentic framework that bridges robust deployment execution and recursive policy self-improvement. During deployment, the Inner Loop decouples high-level reasoning from low-level control through highly composable atomic skills. It employs Vision-Language models for receding-horizon planning and visual reflection, dynamically composing skills to ensure robust error recovery. These skills are executed by specialized flow-matching experts that share a unified VLM backbone, maximizing reusability while mitigating capacity interference. Concurrently, the Outer Loop drives automated lifelong learning by autonomously segmenting and verifying deployment rollouts, clustering them to discover atomic skills, and continuously fine-tuning the skill library without human annotations. Evaluations on RoboCasa, BEHAVIOR-1K, and real-world tasks demonstrate the effectiveness of MobiAgent. It outperforms $π_{0.5}$-TA by 22.5 percentage points on BEHAVIOR-1K and enables robust recovery from execution failures. Through autonomous data recycling, success improves from 7.50% to 27.50% on RoboCasa and from 32.5% to 57.5% on Astribot S1.

Many reinforcement-learning (RL) problems are non-stationary yet structured and can be decomposed into phases, each with its own transition probabilities and reward functions. When the phase sequence is known, the common solution augments the state with information to satisfy the Markovian property and applies standard RL techniques. However, prior work finds that the multi-policy approach for different phases can outperform a single state-augmented policy shared among the phases, for reasons that remain unclear. In this work, we first show that the shared policy can theoretically achieve performance of any multi-policy solution. However, whether a multi-policy solution can perform better than the corresponding single shared policy in practice depends on function approximation, learning and optimization processes, as well as, for multi-policy solutions, the sample efficiency and loss of continuity from one policy to another. We propose a regime-based phase decomposition method to identify which policy can provide better performance. The method is based on consideration of the duration of transient system dynamics relative to the duration of the quasi-stationary period. Numerical experiments are conducted with different non-stationary RL problems to validate our four major hypotheses: (a) longer phase durations favor multi-policies, (b) the heterogeneity between phases increases the burden on single policy, (c) multi-policies need sufficient data for each phase, and (d) environment-specific transition dynamics between phases can affect which policy is preferable.

Accurate colon segmentation from CT images is essential for colorectal disease analysis, yet deep learning based methods often produce disconnected predictions due to complex anatomy. This study introduces a three-stage, topology-preserving segmentation pipeline to address this issue. The first stage performs initial deep learning-based segmentation, followed by centreline bridging to reconnect disjoint regions and a reconstruction stage to refine continuity. Evaluations on TotalSegmentator and RAOS datasets using overlap, distance and topology-based metrics demonstrate improved structural consistency while maintaining segmentation accuracy. The proposed method enhances topological integrity, enabling more reliable colon segmentation for clinical and research applications.

K-fold cross-validation (CV) is widely used as evidence of out-of-sample performance, although folds are neither independent experiments nor equally informative under heterogeneous data. Cross Upper-Bound Validation (CUBV) replaces point-wise CV accuracy by conservative upper bounds on true risk. Here we generalise CUBV through a single exponential framework in which the moment-generating function of the generalisation gap is controlled by a cumulant envelope gamma(lambda). This yields a family of risk bounds covering Hoeffding-, Bernstein-, dependency-aware, PAC-Bayesian, and heterogeneous source-fusion settings. For K-fold CV, dependence between fold-wise gaps is modelled through a joint sub-Gaussian proxy matrix. Under equicorrelation, this gives an effective number of folds, Keff = K/[1+(K-1)rho], showing that increasing K does not necessarily increase statistical evidence when folds are strongly dependent. The framework is also extended to posterior distributions over predictors and weighted multi-source fusion, where weights are selected by minimising an upper bound on future risk rather than empirical error alone. Experiments with trained linear classifiers on heterogeneous multimodal Gaussian mixtures compare K-fold CV with full-sample resubstitution plus risk correction. Bounds are evaluated by coverage and tightness. In low-dimensional small-sample settings, K-fold partitioning can increase uncertainty because individual folds under-represent minority modes, while corrected resubstitution can remain valid and tighter; this effect disappears as sample size increases. Overall, gamma-CUBV separates observed performance, uncertainty, dependence, model complexity, and confidence into explicit terms, providing a unified route from CV scores to risk statements and a principled validation criterion for heterogeneous small-sample applications such as neuroimaging.

Neural quantum states based on modern deep learning architectures have emerged as powerful representations for quantum many-body systems. In particular, Transformer-based neural quantum states provide expressive models capable of capturing long-range correlations, and their empirical generalization performance has recently been demonstrated. However, a theoretical understanding of their generalization behavior remains largely unexplored. In this paper, we develop a theoretical framework to analyze the generalization properties of Transformer-based neural quantum states under in-context learning. We establish a rigorous inference-time generalization error bound in terms of mean squared error (MSE), showing that the pointwise prediction error decreases inversely with both the number of in-context examples and the depth of the Transformer. We further show that the Transformer depth required to achieve this guarantee scales only linearly with the system size--namely, the number of particles in continuous systems or the number of qudits in discrete systems. Building on this result, we extend our analysis to full quantum states formulated as rank-one density operators, and derive MSE-based generalization bounds over both continuous and discrete domains under physical constraints. Finally, numerical simulations corroborate our theoretical analysis.

Post-training with verifiable rewards can induce reward hacking, motivating the use of monitors within the training objective rather than solely for offline auditing. We show that a low monitor readout does not identify whether such an intervention controls behavior. In a code-generation environment whose dominant exploit is available at the start of the reasoning trace, we train policies against three monitors that pass the same offline gate: an in-domain activation probe and two penalties conditioned on how early the policy commits to its own final answer. The probe score is at its numerical floor from the first recorded training step, and the trained-score median is zero for every prefix-trained run at the endpoint. These readouts estimate different quantities, and we do not compare their scales; within each monitor family, however, low values do not establish behavioral control. Within one fixed configuration, prefix-trained runs with the same zero-median trained score range, by seed alone, from a mixed regime with a low hacking share to near-pure reward hacking. All probe runs reach the hacking regime, but their floor-level readout reflects a mismatch between the position where the probe was validated and the position where it was read during training, not a second instance of this ambiguity. Text-level analysis identifies a prefix failure mode: generic planning and filler shells postpone the exploit past the cut without eliminating it from the final output. Low measured commitment therefore does not distinguish a low hacking share from delayed commitment to the exploit. Offline discrimination and low monitor-aligned readouts are insufficient evidence of behavioral control; an out-of-band behavioral check is required. We characterize the endpoint readout, not its evolution. Code is available at https://github.com/zhezhou1106/spoof-cost.

Drug-target affinity (DTA) prediction is widely used to prioritize candidate compounds before costly experimental screening. DTA models are often compared under a single data split, even though deployment may require extrapolation to new chemical series, new protein targets, or both. We ask whether the distribution shift used for evaluation changes which architecture appears best. We curate 718,800 unique drug-protein pairs from the ChEMBL and BindingDB datasets. We compare a Morgan-fingerprint + protein-CNN baseline with 12 controlled architectures that combine four drug representations with three ESM-2 interaction modes. Mean validation RMSE increases from 0.950 and 0.945 under scaffold and fingerprint-cluster OOD to 1.299 and 1.321 under protein-cluster and dual OOD. Model rankings are similar across the two chemical shifts (tau = 0.79), but agreement with scaffold OOD falls under protein OOD (tau = 0.39) and reverses under dual OOD (tau = -0.55). Held-out evaluation, repeated seeds, group-aware bootstrap analysis, and a size-matched control support the same conclusion: architecture selection depends on the form of extrapolation, not only on average error or training-set size. DTA benchmarks should therefore match the chemical and target shifts expected at deployment.

Recent progress in multimodal, high-dimensional learning has enabled foundation models to process heterogeneous, large-scale data. However, at test time, acquiring all features or modalities can be prohibitively costly and often redundant. Sequentially selecting informative modalities is therefore critical, yet challenging when the downstream task or prediction target is unknown. To this end, we introduce ECHO-$k$, a task-agnostic and self-supervised learning principle for modality acquisition: we use a deep model's internal pretrained representations (e.g., from a foundation model) as proxy targets that summarize cross-modal information. We provide theoretical guarantees in a stylized linear setting that motivate a reinforcement learning (RL) policy for sequential modality selection. Across task-agnostic and label-free acquisition baselines, ECHO-$k$ consistently improves budgeted downstream performance across diverse foundation-model backends. Our method provides a principled route to cost-aware test-time deployment, with implications for any multimodal system where measurements are expensive or time-constrained, and downstream tasks unknown a priori.

Causal representation learning for time-series data aims to identify latent states and their causal relations from observations. In this setting, an important challenge is to model both lagged causal relations across observation intervals and faster causal effects that appear as instantaneous relations within an interval, while accounting for nonstationarity in time-series data. However, methods that jointly handle these causal relations and nonstationarity remain limited. To address this gap, we establish sufficient conditions for identifying latent states up to component permutation and component-wise invertible transformations, and their instantaneous and lagged causal structures up to the same permutation, using an observed auxiliary variable, such as time or a condition label, associated with changes in transition-noise distributions. Based on these results, we propose iCReN, a framework that uses contrastive learning with discrete or continuous auxiliary variables to learn latent representations and estimate their instantaneous and lagged causal structures. Experiments demonstrate accurate recovery of latent states and both instantaneous and lagged causal structures on synthetic data and the utility of the learned representations for downstream forecasting on real-world data.

LLM agents increasingly screen tool outputs with small prompt-injection detectors, and teams choose among detectors by their scores on public benchmarks. We ask whether those scores predict how a detector behaves inside an agent. We replay the ground-truth tool calls of two agent benchmarks, AgentDojo and tau-bench, without an LLM to obtain tool outputs that are benign by construction, label injected outputs by differential replay, and evaluate fifteen detectors, including Meta's Prompt Guard 2, and two task-aware LLM judges on these outputs and on the BIPIA benchmark. Detection rankings transfer poorly between benchmarks: the best detector on BIPIA catches 2% of AgentDojo injections at a 1% false-positive rate, and a detector that catches 72% of AgentDojo injections catches 15% on tau-bench. False-positive rates on tool outputs, which range from none to over 90%, do transfer between the two agent benchmarks. Where training data is public, the form of the training inputs explains the results. The BIPIA leader was trained on full BIPIA inputs, but having seen InjecAgent's attack strings as short prompts does not help it find them inside tool outputs; the best detector on both agent benchmarks shares no data with any benchmark and was trained on agent-style inputs. Evaluations meant to inform deployment should use the agent's own tool outputs, report detection at a low false-positive rate, and audit what the detector was trained on.

A targeted adversarial perturbation can drive a vision-language model's (VLM's) teacher-forced training loss for a fixed target caption to near zero, yet the same model, allowed to generate freely, produces the original, correct description with no trace of the target. We call this dissociation the train/inference gap, and give it a precise mechanistic account on Qwen2.5-VL-7B-Instruct using a controlled two-stage PGD attack on 200 held-out COCO images. First, we show that image-level pixel statistics, including a correctly re-implemented, texture-based attackability measure from the CNN robustness literature, have essentially no predictive power over which images are corrupted (best predictor r=-0.050, p=0.484; ridge regression R^2=0.069). Second, using the logit lens, we localise the gap to a single autoregressive step: the rank of the target token, conditioned on the correct first token already being generated, is fixed at exactly 3,488 out of 152,064 vocabulary entries for every image and every condition, with zero variance. Third, tracking target-token rank across all 28 LLM decoder layers reveals that the visual encoder corrupts every image's representation by a comparable margin regardless of eventual outcome, but the language model decoder then differentially arbitrates: amplifying the corrupted signal for susceptible images and actively suppressing it, past its clean-image baseline, for resistant ones (p<0.001, rank-biserial r=0.579). A linear probe on the merger hidden state separates these two outcomes with AUC=0.858, though we flag a circularity concern in this estimate. Together these results argue that adversarial robustness in autoregressive VLMs is substantially a property of the language decoder's prior, not the visual encoder, with direct implications for where faithfulness evaluations and defenses for deployed VLM systems should be targeted.

Molecular optical absorption spectroscopy provides a direct probe of electronic structure and is widely used for molecular identification, interpretation of photophysical behaviour, and planning of spectroscopy experiments. Calculating the absorption spectra using first-principle excited-state methods, however, is computationally demanding, at least compared to ground-state calculations, which limits their routine application across large molecular sets. Machine-learning (ML) surrogates can reduce this cost and allow rapid spectral prediction. However, their performance depends strongly on how molecular information is represented. Here, we compare using the ground-state electron density versus the molecular geometry as inputs to a ML model for predicting absorption spectra, for a training set of 6874 molecules selected from the QM7 dataset. For each of these molecules, the density was calculated using density functional theory (DFT) and the absorption spectrum was calculated using linear-response (LR) time-dependent DFT (TDDFT). Utilizing the ground-state density as the input to the ML model is motivated by the Hohenberg-Kohn and Runge-Gross theorems, and the fact that the ground-state density encodes information about bonding, charge localisation, and electronic delocalisation. Hence, it may be a more judicious starting point for the ML model compared to the geometry, as it effectively decouples the chemistry of the ground-state. The question we test is whether the benefits of using the density outweigh the (notprohibitive) penalty of requiring an additional single-point DFT calculation for the density. We find that the density-based convolutional neural network achieves a validation correlation of 0.9926, compared with 0.9795 for the best geometry-based graph model, reducing the residual decorrelation, by approximately 64%.

Online data selection has demonstrated substantial efficiency gains for LLM pretraining by training on the most valuable candidates within each batch. Since a candidate's value is realized through its effective model update, principled selection should account for the optimizer step, which reshapes the raw gradient before it updates model parameters. We formalize this optimizer-aware selection paradigm as OptiSelect and present the first systematic study of how the optimizer shapes data selection. Our theory establishes a selection gain principle in which the advantage of online selection is governed by the discriminability of the optimizer-induced utility scores. We prove that sign-based and polar-tangential preconditioners of Lion and Muon would suffer from a discriminability collapse which caps attainable gains from OptiSelect, whereas diagonal-adaptive optimizers such as AdamW and Sophia admit strictly better upper bounds. The proposed principle also yields a quantitative derivation of the optimal candidate oversampling ratio. Pretraining experiments on 124M and 720M models are consistent with our theoretical analysis and show that AdamW's diagonal-adaptive scoring geometry remains the strongest scoring geometry even with Muon as optimizer. We further demonstrate that OptiSelect retains its benefits under data rephrasing, a technique used in modern data processing pipelines. Our findings provide theoretical foundations and practical guidance for co-designing optimizers and data selection in LLM pretraining.

Efficient request scheduling is increasingly important for reducing completion time in large language model (LLM) serving. Size-based policies such as Shortest Job First prioritize shorter requests, but output lengths are unknown before generation, so practical schedulers rely on predicted lengths. We introduce JIL, an attack on prediction-based LLM schedulers that manipulates the scheduling signal to obtain higher priority and reduce completion time. Using TRAIL as a case study, JIL optimizes an adversarial suffix that causes a lightweight output-length probe to underestimate a request's length. We evaluate JIL on two datasets and four LLMs across varied request profiles and deployment configurations. JIL reduces predicted output lengths by up to 83.4 percent, and adversarial requests complete up to 1.53 times faster on average in end-to-end serving experiments. The reduction in predicted length is substantially larger than the change in actual output length, revealing a mismatch between the scheduler's estimate and the request's realized size. Response utility varies across models and tasks, exposing a trade-off between scheduling advantage and response quality. We also evaluate scheduler-side defenses and find that grouping length predictions into coarse intervals reduces JIL's scheduling advantage and mitigates delays to benign requests.

Suspicion is an important, yet elusive concept in anti-money laundering and counter-terrorist financing (AML/CFT), which allows for intervention below the threshold of proof. In its traditional form, suspicion can be understood as a situated legal judgement by human actors within identifiable jurisdictions. It is argued that this understanding is no longer adequate. As artificial intelligence (AI) becomes an integral part of financial surveillance, suspicion is increasingly produced through data-driven processes. This transformation is epistemic, but also spatial. Since AI-driven financial surveillance operates through transnational data infrastructures, regulatory reach is less a matter of where conduct occurs than a question of whether such conduct becomes visible within data systems. This article develops the concept of algorithmic extraterritoriality, understood as a form of regulatory power mediated by data infrastructures rather than formal assertions of jurisdiction. Moreover, since individuals are increasingly constituted as datafied subjects of suspicion, they are rendered governable through dispersed and opaque processes of evaluation. This constitutes a challenge for accountability and contestability because suspicion becomes more difficult to locate, explain or contest.

Multimodal large language models (MLLMs) have shown strong visual reasoning abilities, but knowledge-intensive visual question answering often requires external textual evidence beyond the image and the model's parametric knowledge. Existing multimodal RAG systems commonly rely on Top-$K$ retrieval or reranking, which may return redundant passages and provide limited control over whether an answer update is sufficiently supported by the retrieved evidence. We propose \textit{CLIMB}, a training-free inference-time framework for multimodal RAG. CLIMB first constructs a compact complementary evidence pool using an MMR-style objective that balances query relevance and passage-level redundancy. It then performs confidence-controlled refinement within this fixed pool: an R/E/C critic scores passages by relevance, evidence specificity, and cross-modal alignment, while an evidence-grounded confidence estimator accepts an updated answer only when the estimated confidence increases. This design provides a simple stopping criterion and reduces unnecessary refinement without modifying the underlying retriever or MLLM. Experiments on Encyclopedic-VQA and InfoSeek show that CLIMB consistently improves over retrieval-augmented multimodal baselines. Ablations further indicate that complementary pooling, critic-based scoring, and iterative confidence-controlled refinement each contribute to the final performance.

In this work we formulate ultrasound multistatic recovery from arbitrary transmit sequences as a Bayesian inference problem. To that end, we train a deep generative prior on multistatic data sets to tackle the rank-deficient regime in which classical linear REFoCUS decoders fail. This appproach, which we term Deep Bayesian REFoCUS, outperforms the linear baselines for all regimes of rank-deficiency and noise levels, and regresses to linear decoding when inversion is exact. The model also expresses uncertainty in the null space of the acquisitions, whereas the linear REFoCUS decoders only provide point estimates. Finally, we analyze the impact of distribution shift between simulation and in-vivo acquisitions, showing remarkable generalization ability without any fine-tuning or adaptation.

Epistemic uncertainty should decrease as additional information about the data-generating process (DGP) becomes available to the predictor. Yet, existing graph evidential deep learning (EDL) methods for node classification typically construct epistemic uncertainty from graph-specific properties and evaluate it on downstream tasks such as out-of-distribution detection, which do not test its reducibility as information about the DGP increases. To make reducibility directly testable, we introduce a statistical framework for studying epistemic uncertainty under information growth. Our framework specifies an information-growth experimental protocol and a consistency criterion for epistemic predictors, while using projective graph DGPs to ensure that growing graphs, which in general need not provide increasing information about the same DGP, constitute coherent observations of the same underlying process. We show that EDL methods do not explicitly estimate data uncertainty arising from a single finite graph observation and instead regulate epistemic uncertainty through model hyperparameters, precluding consistency, as corroborated by controlled information-growth experiments. As an alternative, we propose graph bootstrap ensembles, capturing both data and procedural uncertainty through graph resampling and randomized training. Under the same experimental protocol, these ensembles exhibit epistemic uncertainty reduction beyond standard deep ensembles. These findings support bootstrap ensembles as candidate consistent epistemic predictors under information growth.

Consistency models (CMs) have become a leading approach for generating high-quality samples in few steps. However, adding steps can improve or degrade sample quality in ways that are highly sensitive to the schedule and that existing theory does not fully explain. To provide accuracy guarantees and guide CM sampler design, we analyze multistep CM sampling as a composition of noising and approximate denoising operators. Under explicit, verifiable stability assumptions, we derive a non-asymptotic error bound that separates contraction of the initialization error from accumulation of approximation error. The bound assigns distinct roles to the schedule: large early noise levels drive contraction, while small late noise levels control the residual bias. As a corollary, we obtain explicit constants for strongly log-concave and semi-log-concave targets. We further establish a complementary guarantee whose assumptions, one-step accuracy and stability, can be estimated for a given trained model. Experiments show that the contraction and approximation profiles entering our bounds can be reliably measured and closely match the predicted functional forms. Together, these results provide a meaningful convergence theory for multi-step CMs and a practical route to sampler design.

Sequential learning systems often make decisions from accumulated experience while receiving high-dimensional inputs whose distribution may change over time. Instance-Based Learning Theory (IBLT) provides a principled case-based framework for such settings through stored situation-decision-utility instances, partial matching, activation, and blending. IBLT relies on symbolic knowledge representation in dictionary-like formats, but text, images, transaction vectors, and user-item histories often require learned similarity rather than hand-specified matching rules. In this paper, we introduce AIBL (Augmented Instance-Based Learning), an instance-learning model formulated in a learned vector space for high- dimensional sequential data. AIBL generalizes symbolic situation matching to neural embedding similarity while retaining instance storage, activation- weighted retrieval, and utility blending. The AIBL model organizes memory into active, forgotten, and surprise stores. Surprise memory separates weakly matched, possible out-of-distribution, or corner-case observations from active memory, reducing forced fitting to the nearest available cases. An observation-driven graduation algorithm promotes recurring surprise instances to active memory, allowing the memory to incorporate repeated novel patterns that may arise under concept drift. We evaluate the same implementation on five machine learning tasks and three controlled simulation tasks, comparing AIBL with classical IBLT variants and task-specific baselines where appropriate. AIBL improves accuracy by 6 to 17 percentage points. The results show where vector-space retrieval improves over symbolic matching and how the added memory mechanisms govern novelty detection, cold-start handling, drift adaptation, and reward learning under the tested protocols.

Contrastive representation learning is increasingly used to recover low-dimensional structure from neural recordings, but its output is typically validated by decoding accuracy rather than by the geometry of the manifold it produces. We apply CEBRA to EEG recorded from dyads in conversation, and analyze the resulting embedding, which training constrains to the 2D sphere. Labels describing the dyads, including the absolute difference between partners' autism-quotient scores, decode well above chance (0.77 against a 0.55 majority baseline for binary AQ magnitude; 0.44 against 0.25 for the six-class $|Δ$AQ$|$ partition). However, the two permutation controls have notable differences in results: permuting labels over a frozen embedding yields p = 0.001, whereas retraining the encoder under each permutation yields p = 0.50. Only the latter tests the label rather than the geometry. Consistent with this, spherical mixture structure and per-class dispersion track identity rather than autism trait differences in dyads; frequency-band and non-oscillatory activity ablation controls do not change the results. However, participant-level model does separate from its identity-aware null (p = 0.0099) while speaker-versus-listener role analysis performs at chance in the same embedding, indicating a manifold organized by individual -- and, in contrast with current neurolinguistics models, almost invariant to speaking vs. listening. Based on these results, we suggest that retraining-based nulls should be the default for grouped-data contrastive embeddings.

Forest point cloud segmentation is fundamental for fine-grained 3D forest scene understanding, yet remains challenging due to irregular tree structures, severe occlusions, density variations, and ambiguous instance boundaries. Recent query-based forest segmentation methods have shown promise for unified semantic and instance prediction, but they still insufficiently exploit forest-specific spatial structure and account for boundary uncertainty. In this paper, we propose ForestQuery, a boundary-aware and spatially anchored query learning framework for unified forest point cloud segmentation. ForestQuery enhances instance and semantic query learning through two complementary designs. Specifically, boundary uncertainty is explicitly modeled to guide reliable instance query construction and modulate query optimization through adaptive loss reweighting. Meanwhile, spatially anchored semantic query enhancement (SA-SQE) introduces learnable 3D anchors encoding forest vertical stratification priors to enrich semantic queries with explicit spatial references. We evaluate ForestQuery on multiple public forest point cloud benchmarks and a self-collected annotated real-world dataset. Extensive experiments demonstrate consistent improvements in both individual-tree segmentation and semantic segmentation across diverse forest scenes. Code and data are publicly available at https://zhan994.github.io/ForestQuery

To deploy deep neural networks on edge hardware, highly efficient inference schemes are necessary that retain high accuracy. This work presents W16A16, a high precision (16-bit), fast speed, low energy quantization method. On a widely applied microcontroller architecture Armv7E-M, our proposed approach achieves faster speed and lower energy consumption on layer- and model-level compared to alternative quantization schemes. We analyze the architecture of Armv7E-M, explain the underlying principles behind the performance advantages of 16-bit approaches, and evaluate the empiric quantization errors for regression and classification tasks, as well as empiric time- and energy consumption in MCU deployment. We observe ca.\ 10 times lower quantization errors compared to 8-bit quantization schemes while achieving similar or better inference times and energy consumption.

Bayesian data assimilation combines model forecasts with noisy observations, but sampling high-dimensional, non-Gaussian posteriors remains challenging. We introduce an observation-interpolant framework that turns pretrained stochastic interpolant, flow matching, and diffusion models into posterior samplers without retraining. Conditioning the interpolant path on observations yields a shared likelihood-score correction to the drift or velocity, unifying stochastic and deterministic posterior sampling. The resulting SDEs and ODEs sample the exact posterior when the intermediate likelihood score is known. For practical computation, we approximate this score using a closed-form Gaussian surrogate with a bias-corrected mean and covariance inflated by the model's source covariance. Jacobian-free and ensemble-shared approximations make the method tractable in high dimensions. We evaluate the framework on linear-Gaussian dynamics, stochastic two-dimensional Navier-Stokes, and urban airflow with up to $O(10^4)$ degrees of freedom.

Long-horizon tasks with sparse rewards pose an exploration bottleneck for goal-conditioned reinforcement learning: a policy started from the initial state rarely reaches the goal and receives no learning signal. Reference motions, hand-designed curricula, and shaped rewards supply this signal but require demonstrations or task-specific engineering; automatic start-state and goal curricula avoid this but typically expand from one side only, so the full distance to the target must be covered from that side. We propose the Bidirectional Voronoi-biased Exploration curriculum for Reinforcement learning (BVER), which expands from both ends at once. Inspired by bidirectional RRT planning, BVER grows start states outward from the goal and goals outward from the initial state distribution, biases both toward unexplored task space, and steers them toward each other, training one goal-conditioned policy on both. On point-mass mazes, quadrupedal box climbing, and robot-arm ring-on-peg transfer, BVER learns faster than all compared reference-free curricula. On box climbing, it reaches 95% success on a 0.4 m box in roughly 65% fewer iterations than the best of them, is the only one of them to learn to climb a 0.7 m box, and yields a policy robust to start, goal, and yaw variation. Without a demonstration, it approaches the sample efficiency of reference-based curricula on the 0.4 m box and on ring-on-peg transfer. Ablations show that expanding from both ends outperforms either direction alone.

Few-step neural text-to-speech models often rely on short- ened diffusion or flow-matching schedules, or on distillation from pretrained multi-step teachers. To avoid these depen- dencies, we present DriftTTS, a few-step mel-spectrogram generator trained without a generative teacher, distillation, or adversarial discrimination. DriftTTS uses a distribution- matching drift objective in a mel-domain feature space defined by raw mels and a frozen masked-autoencoder encoder pretrained on the same LJSpeech training split. On-policy rollout trains the decoder on its own interme- diate states and supports inference up to the trained roll- out depth. On LJSpeech, DriftTTS at NFE=4 achieves 3.87 dB MCD and 3.7% WER, compared with 3.85 dB and 3.4% for Matcha-TTS. In a fully paired blind listen- ing test, DriftTTS obtains 4.18 MOS, compared with 3.96 for Matcha-TTS and 4.22 for ground truth. These results demonstrate competitive few-step synthesis without a pre- trained generative teacher. Code can be found at https: //github.com/BASHLab/driftTTS.git

Vision language models (VLMs) incur substantial inference cost because every visual token is processed by the attention and MLP projections of every decoder layer, even when token-specific visual computation is unnecessary at many depths. We introduce Patch-to-Prune (P2P), inspired by Mechanistic Interpretability, a training-free framework that converts activation patching from a diagnostic tool into an inference-time computation bypass. P2P performs validation-guided forward and backward layer sweeps to identify decoder regions whose visual-token projection outputs can be replaced by fixed neutral proxy activation vectors within a user-specified accuracy tolerance. Unlike conventional token-pruning methods, P2P preserves the sequence length, token order, positional information, attention mask, and residual pathways, thereby pruning computation without removing tokens or modifying the pretrained model weights. We evaluate P2P on four VLMs from the Qwen2.5-VL and LLaVA families across seven multi-modal benchmarks using mutually disjoint calibration, validation, and test partitions. P2P at a 3% tolerance retains around 94% of dense accuracy while reducing FLOPs by 55%. Beyond these efficiency gains, our layer-wise analysis suggests that visual processing in VLMs is non-uniformly distributed across decoder depth: early and late layers often require little token-specific visual computation, whereas intermediate layers appear to perform most task-relevant visual integration, enabling later reasoning to rely largely on visual information already embedded in shared residual and textual representations. This makes P2P both an efficient inference framework and a causal lens into visual information processing in VLMs.

Typed decision models return choices or distributions over answer options supplied at request time. Accuracy with complete options does not establish whether a model recognizes that a reference answer is missing or avoids rejecting valid candidates. We present a paired candidate-coverage benchmark protocol and an initial evaluation of Laya and Jev across AG News, DBpedia, Emotion, and TREC. The models receive identical frozen texts and requests: 300 calibration and 589 test texts yield 23,932 predictions per model. Present/absent pairs match ordinary candidate count, and name variants preserve descriptions, members, and order. Native rejection behavior differs sharply: at five TREC candidates with natural names, Laya detects 97.2% of missing-answer cases but falsely rejects 69.7% of present controls; Jev's rates are 24.8% and 0.0%. Calibration-only none-score thresholds change these rates to 33.9%/3.7% and 45.0%/1.8%, respectively. On DBpedia, Jev's high coverage-score AUROC supports a stronger operating point, whereas both models have weak complete-set accuracy on Emotion. Competence-conditioned analysis, probability-precision sensitivity, and interface audits show why classification, score ranking, and rejection policies need separate measurement. This initial benchmark is descriptive and limited to reference-label omission; it does not establish natural out-of-scope generalization, causal mechanisms, or a new rejection method.

Attack Path (AP) modeling is fundamental to cybersecurity analysis, where the Planning Domain Definition Language (PDDL) has been widely adopted to encode APs into formal and machine-verifiable representations for automated reasoning about vulnerability exploitation, attack progression, and their potential impacts. However, existing AP modeling approaches largely rely on expert-driven manual construction, limiting their scalability and ability to keep pace with rapidly evolving cyber threats. Large language models (LLMs) are promising candidates, as their extensive pre-trained knowledge and reasoning capabilities enable them to interpret and transform threat intelligence into formal representations. In this paper, we propose \textbf{CVE2AP}, an LLM-based approach for automatically generating PDDL-encoded attack paths from natural language CVE (Common Vulnerability Exposure) descriptions. CVE2AP leverages structured prompting and incorporates an error-feedback mechanism that iteratively refines the generated paths using planner-reported syntactic and solvability errors. We conduct a systematic empirical evaluation across multiple LLMs and generation configurations, assessing generation quality across syntactic, solvability and semantic dimensions, together with token consumption and generation time. The results demonstrate that CVE2AP effectively generates high-quality PDDL-encoded attack paths, achieving up to 86.9\% syntax correctness, 78.6\% solvability, and 93.1\% semantic correctness under LLM-as-expert evaluation, while \texttt{GPT-5.5} offers the best quality-cost trade-off and error feedback yields the most consistent quality improvement.

Physics-Informed Neural Networks (PINNs) typically exhibit spectral bias, where some frequencies of the target function converge more slowly than others. In this work, we analyze the training dynamics of Fourier Feature PINNs in the Neural Tangent Kernel regime to address this limitation. We derive an explicit evolution equation to estimate the residual error in the frequency domain, demonstrating that the convergence rate of specific frequencies is primarily governed by the product of the differential operator's symbol and the spectral density of the initialization weights. Leveraging this theoretical insight, we propose an informative initialization strategy that tailors the initial weight distribution to the specific PDE being solved. With this method, we can diminish the operator-induced spectral bias, balancing the convergence rates across the frequency spectrum and achieving better prediction accuracy. Numerical experiments on linear and nonlinear partial differential equations confirm that this initialization strategy improves learning dynamics and approximation accuracy across frequencies compared to standard initialization methods, with no additional training cost.

Training long-horizon agents to solve complex tasks requires effective supervision over extended interaction sequences. However, sparse terminal rewards obscure intermediate contributions, while on-policy distillation can lose informative teacher guidance as student-generated histories grow. To address this problem, we introduce SCAD, which organizes interactions into planning and bounded subtask execution, distills execution in local contexts, and refines planning credit through cross-rollout subtask prefix trees, with planning receiving full terminal credit and execution receiving positive terminal credit and teacher guidance. Across all evaluated benchmarks, SCAD improves macro-average accuracy over the strongest training baseline by 4.48 percentage points for text tasks and 4.19 points for multimodal tasks. SCAD effectively combines outcome-based credit assignment with teacher-guided distillation to improve planning and execution in long-horizon agents.

Deep reinforcement-learning policies for order execution can vary substantially across training seeds, so apparent architectural gains may reflect favourable training realisations rather than reproducible properties of the architecture. We evaluate vanilla Double Deep Q-Learning (DDQL), K-means-partitioned mixtures of DDQL experts at $K \in \{2, 4, 8\}$, and dense networks parameter-matched to the $K{=}4$ and $K{=}8$ expert budgets on 5-minute mean-aggregated BTC/USDT limit order book data from Binance. No learned configuration significantly improves mean implementation shortfall over DDQL. Under the reported specification, all have higher mean shortfall than TWAP (0.39 bps) and immediate liquidation (0.21 bps) in an environment whose frictionless replay and terminal-urgency penalty make early liquidation nearly costless; 11/100 vanilla-DDQL runs, versus none in either MoE $K{\geq}4$ arm, converge to a policy that waits until forced liquidation. We then decompose this specification on a device-matched baseline. Annealed exploration alone eliminates observed collapses (12/100 to 0/100; exact McNemar $p{=}4.9{\times}10^{-4}$), matching the elimination under expert partitioning. Combining annealed exploration with the aligned reward restores collapse in 19/30 runs; with all three specification changes, it rises to 48/100. In this environment, expert partitioning is unnecessary to suppress collapse and appears to mask a training-specification failure rather than confer an intrinsic performance benefit. No MoE $K{=}8$ run collapses under any of the six specifications tested. Across-seed dispersion is lowest at $K{=}8$ but non-monotone and not robust to family-wise adjustment, while within-policy tail risk worsens monotonically with $K$. The apparent attribution of the failure mode reverses between 30 and 100 seeds, illustrating the importance of repeated-seed evaluation.

We understand little about how capabilities acquired in one language carry over to another, or what governs this transfer: evaluations rely on incomparable, saturation-prone datasets and rarely examine its determinants jointly. Identifying what predicts transfer would let us avoid exhaustive evaluation across all language pairs and let developers target the factors that limit performance in low-resource languages. To evaluate cross-lingual capability transfer, we introduce Multilingual GSM-Symbolic, an extensible multilingual mathematical dataset covering 30,000 item-matched question-answer pairs and spanning 15 languages. It utilises symbolic templates to prevent overfitting and ensure generalisation by allowing generation of millions of high-quality variations from a single sample. Using Multilingual GSM-Symbolic, we quantify the largest determinants of capability as model size ($β= 1.77$), language resource level ($β= 0.77$), reasoning ($β= 0.67$) and typological distance ($β= -0.25$). This joint estimation allows these determinants to be expressed in terms of one another: a 32B model evaluated in Marathi performs like a 10B model in English. Our findings have important implications for model developers, showing that model size and reasoning narrow the performance gap between low- and high-resource languages ($β= -0.27$ and $β= -0.20$, respectively), while similar levers have little or no effect on typologically distant languages. Overall, our analysis framework explains 92% of between-language variation, but only 23% of the model-by-language variation, and predicts a model's performance on an unseen language within 6.0pp (r=.96). Incorporating measurements from just 10 templates in the target language reduces this to 4.19pp, enabling reasonable estimates of performance with little or no downstream dataset.

Neural surrogate models for Partial Differential Equations (PDEs) on unstructured 3D geometries are often limited by poor generalization and the high cost of generating large-scale training datasets. Consequently, pre-training on massive datasets of related PDE dynamics has emerged as a critical alternative to enhance the robustness and scalability of these models. However, this strategy is neither compute- nor data-efficient, as it relies on massive pre-computed data that is very costly to generate. In this work, we introduce a disk-data-free pre-training framework tailored to both steady-state and transient regimes. For steady-state problems, we propose a geometry-driven strategy that leverages intrinsic shape descriptors to learn representations of complex 3D domains. For transient problems, we introduce a physics-driven approach based on online generation of synthetic PDE data, enabling scalable pre-training without reliance on expensive datasets. Across multiple experiments, our approach achieves faster convergence, greater data efficiency, and higher accuracy during fine-tuning, particularly under realistic low-data regimes. This methodology provides a practical pathway toward data-efficient neural emulators for large-scale simulations.

Critic-free reinforcement fine-tuning (RFT) for agentic large language models is often done through GRPO-style methods, which compute a group baseline over repeated rollouts to reduce target variance. However, this setup is ill-suited to agents acting in stateful environments such as live services or security sandboxes, where repeated rollouts are impractical to obtain and aggressive updates entrench the noise of long, sparsely verified trajectories. We propose \textit{Follow the Winners} (FTW), a critic-free policy-learning algorithm that adapts the cross-entropy method to RFT, replacing group rollouts with an ordinal filter on replay-buffer samples that yields polynomial concentration in the order statistic of returns. We derive FTW through a control-as-inference lens, which also recovers GRPO and DPO as specific modelling choices, identifying GRPO as risk-neutral while DPO and FTW share a bounded risk-seeking offset that FTW controls. We identify this offset as an inherent trade-off of variance reduction through ordinal filters on samples, whereas a critic model induces a different trade-off between bias and variance. Scaled to agentic LLM post-training, FTW matches GRPO and PPO on Sokoban and Search-R1 baselines, showing a viable trade-off from a value model or group rollouts to CPU memory.

Large Language Model (LLM) agents are increasingly deployed in high-stakes settings such as industrial maintenance and equipment fault troubleshooting, where workers occupy a variety of roles. A capable agent must therefore act in a way that is calibrated to user's role: taking actions and providing information that respect the role's knowledge and capability boundaries. Unlike coding, where mistakes are usually recoverable, agent responses in these settings are enacted on physical equipment, and can therefore cause irreversible equipment damage, production loss, or personnel harm. Existing benchmarks, however, largely overlook the need for agents to infer what a role intends and acting only through tools that role may legitimately use, a capability which we term Perspective Awareness. To this end, we introduce ReFract, a benchmark of 150 expert-validated entries in which an agent must act differently in response to the same query depending on user's role. Entries of ReFract are grounded in anonymized queries from domain support conversations, against which we construct Text World Models that simulate the agent's operating environments and assemble perspective-aware action trajectories. State-of-the-art LLMs solve at most 69% of the tasks with more than 50% of their trajectories contain attempts of taking perspective-violating actions. ReFract exposes perspective awareness as a distinct, largely unsolved axis of agent evaluation and motivates agents that calibrate not just how to act, but for whom.

Imitation learning for contact-rich manipulation requires high-quality expert data that is expensive to obtain. This makes learning a sample-efficient policy a key issue. To address this, we propose VISTA, a workspace-level equivariant visuotactile diffusion policy for data-efficient contact-rich imitation learning. VISTA projects visual and tactile observations into spherical tokens, injects tactile contact cues into visual spherical directions through permutation-equivariant spherical fusion, and rotates the fused harmonic representation using the end-effector orientation. The resulting representation conditions an equivariant diffusion policy to predict spatially consistent actions. Extensive experiments in both simulation and real-world robotic settings show that VISTA substantially improves data efficiency over strong visuotactile imitation learning baselines. Project website: https://vista-paper.github.io/

We consider a distributed learning task with agents that have correlated data. Specifically, the label of an agent depends on the input of other agents for the same sample, and these inputs are also correlated. Correlated data is the reality when agents share the same environment. Existing decentralized methods, such as federated learning, ignore the structure of the problem and perform poorly on correlated data. On the other hand, centralized approaches are infeasible due to privacy and communication constraints. We introduce cordial (correlated and distributed) learning to address this gap by sharing only low-dimensional outputs between the agents while training local models to extract informative signals from peers. This distributed learning induces a game in which the loss function of each agent depends on the models of others. Assuming a linear model, we prove that cordial learning converges with probability one to a globally optimal solution, despite the nonconvex global objective. Experiments on structured multi-digit MNIST tasks demonstrate that cordial learning remains highly effective even in highly nonlinear settings.

Large language models are increasingly used where small syntactic errors matter, yet character-level reasoning is still evaluated mostly through isolated probes and aggregate accuracy. We introduce SyntaxBench, a diagnostic benchmark and statistical evaluation framework for character-level reasoning. It contains five core tasks, character counting, letter containment, palindrome detection, edit distance, and longest-string selection, plus index_to_span, a harder substring-extraction stress test. The five core tasks use paired English and character-length-matched random-string inputs. index_to_span documents share a 200-500 word band and are not character-length matched. All six tasks use zero-, one-, and four-shot prompts. We evaluate eight open-weight models from 2B to 32B parameters across 11 reasoning-mode configurations. The framework reports exact-match and relaxed accuracy, Cohen's kappa, paired McNemar tests with odds ratios, bootstrap confidence intervals, Kendall's tau, class-conditional metrics, tokenization analysis, and multiple-comparison-corrected tests. Three findings stand out. First, tokenization shapes accuracy: random strings are more character-visible than English strings (1.892 vs. 3.169 characters per token), and character-counting accuracy falls as English words occupy more tokens. Second, reasoning mode is not uniformly helpful: Gemma4-31B is nearly unchanged across modes on the near-saturated tasks, while Qwen3.6-27B is worse with thinking on palindrome detection (0.952 non-thinking vs. 0.886 thinking at four-shot). Third, index_to_span remains largely unsolved; the best four-shot exact-match accuracy is 6.75%. Character-level evaluation needs controlled inputs, paired tests, and analyses of tokenization and reasoning mode rather than aggregate accuracy alone.

Diffusion language model (dLLM) compression faces a known challenge because calibration is typically performed on clean, fully visible activations, whereas inference traverses partially masked intermediate states. For low-rank compression, this raises two questions. First, can low-rank optimality still be characterized when approximation quality is measured over trajectory-distributed states, and second, does the choice of calibration states affect mathematical reasoning preservation under compression? We address these questions by formulating a trajectory-aware low-rank objective over corruption levels and masking realizations. To estimate this objective efficiently, we propose Traj-MC, which estimates the trajectory second moment through Monte Carlo sampling and yields exact sampled-state optimality and population consistency. Under matched compression budgets, trajectory-aware calibration improves reconstruction over the generation trajectory and preserves substantially more mathematical reasoning than clean calibration on mathematical reasoning benchmarks. Our results connect trajectory-aware low-rank optimality to the reasoning capability retained after dLLM compression. Our code is available at: https://github.com/Zishan-Shao/traj-mc.git.

The scale of online content makes hate-speech moderation challenging, while Large Language Models (LLMs) enable harmful material to be produced and adapted more easily. Moderation therefore requires efficient classifiers that can accommodate different definitions of hate speech. Recent structured decision models accept natural-language criteria and select among specified answers, raising the question of whether they can meet these requirements without task-specific training. We present HATEDECIDE, an evaluation of six decision-model configurations on four hate-speech datasets against specialized moderation, zero-shot, commercial, and supervised baselines. We examine whether supplying a dataset's definition, or decomposing it into multiple questions, improves classification, and we measure their latency and cost. We find that commercial LLMs significantly outperform all decision models on only one dataset. Supplying definitions changes up to 28\% of predictions without consistently improving classification, and decomposition significantly improves performance in only 20\% of the comparisons. On a diagnostic set of test cases, the best hosted decision model comes within 1.6 macro-F1 points of the best commercial LLM at approximately 97\% lower inference cost. These results identify opportunities for inexpensive moderation, while showing that explicit criteria and additional questions do not reliably improve classification.

Low-rank approximation can require additional matrix--vector products to verify that its error meets a prescribed tolerance. We characterize this certification cost for both relative matrix error and mean-square output error. For a single approximation matrix candidate, we determine the exact dimension-uniform minimax query constant as the allowed failure probability vanishes. Our main result concerns reusing validation responses as the approximation space expands. For a candidate family constructed independently of validation, one batch supports an entire nested path without increasing the query budget with the number of checks. Across \(W\) paths, a concentration bound exploiting shared residual energy yields a \(\sqrt{\log(W+1)}\) dependence. A matching lower bound establishes its optimality for fixed interior error targets and sufficiently small separation gaps. Finally, we compare two uniformly valid certificates on the same dispersed-spectrum family. Optimizing the validation budget within each rule family yields costs of orders \(N^{1/3}\) and \(N^{2/3}\) for validation and construction beyond the true target. Code is available at https://anonymous.4open.science/r/Low-rank-approximation-1275/

Diffusion language models for text-to-speech combine two forms of computation: model depth (parameters) and refinement steps (inference budget). We ask whether they scale equally across capabilities. We train 15 masked-diffusion codec TTS models varying depth (19-133M parameters, 3 seeds) on 2,000 hours of speech and sweep refinement steps T in [1,16] at inference, measuring zero-shot synthesis via ASR word error rate (intelligibility) and speaker verification (identity) on 174 held-out speakers. Against measured floors, refinement closes 86.2% of the intelligibility range but only 46.4% of the identity range - a 1.86x asymmetry robust across multiple error metrics. Retraining at 3x and 6x schedule attenuates but does not reverse this gap (1.84 to 1.36 to 1.23x), because intelligibility saturates with steps while identity continues improving. Best-of-K search recovers speaker identity where refinement fails, with 64.6-79.0% win rates across four independent encoders. Depth and steps are not interchangeable: separable B(d)B(T) fits significantly better (Delta AICc=+69.3) than substitution models. Analysis shows 62% of remaining identity deficit lies in the codec, not the generator. We conclude that refinement and depth target different bottlenecks and should be optimized separately.

Designing effective heuristics for diverse combinatorial optimization problems requires substantial expertise and repeated search. Large language models (LLMs) automate heuristic generation and refinement, but heuristic search typically depends on evaluation feedback from the problem being optimized. Generalizing to new problem definitions using only source-task feedback therefore remains a central challenge. We introduce MECo, an LLM-driven multi-task evolutionary framework for zero-shot cross-problem generalization. MECo maintains task-conditioned heuristic populations and uses a transfer gap based on cross-task population performance to guide their interactions. These interactions enable the transfer and recombination of heuristics. A complementary selection criterion then constructs a compact heuristic set by rewarding each member's additional coverage of source combinations. The selected set is applied to target problems without further search or adaptation. Experiments on 32 problem variants across vehicle routing (VRP) and flexible job-shop scheduling (FJSP) show that MECo achieves the lowest mean costs compared with eight automated heuristic design (AHD) baselines under the same budgets. On out-of-domain problems, it outperforms the strongest baseline in each family. Moreover, integrating the framework of MECo with different AHD methods improves their ID and OOD performance in both families, supporting its effectiveness across different methods.

Trajectory evaluation is essential for improving the reliability of LLM-based agents, but production use makes it expensive to run repeatedly. Modern agents generate long traces containing tool calls, observations, retries, and external outputs, while not all raw tokens are equally useful for diagnosis. We present \textit{LiteTrajEval}, a lightweight architecture for budget-bounded trajectory evaluation. LiteTrajEval derives compact domain-specific rule profiles offline, then preprocesses each trajectory online, marks heuristic failure signals, serializes it under a fixed global budget, and invokes a single rubric-guided LLM judge to produce structured diagnostic reports. Evaluated on public Magentic-One-style and $τ$-bench-style trajectory datasets, LiteTrajEval improves failure-localization alignment with human annotations by roughly 20--35 percentage points on Magentic-One and up to 23 percentage points on $τ$-retail compared with AgentRx, while reducing cost by about 6$\times$ and evaluation time by more than 8$\times$. This solution has also been deployed in our enterprise agentic platform.

Flow- and diffusion-based generative models have recently emerged as flexible and highly efficient forecasting models for dynamical systems. When combined with inference-time guidance, they offer a promising route to high-dimensional non-Gaussian data assimilation (DA), the problem of combining forecasts with observations to estimate latent system states. Existing filters, however, condition on a fixed history and assimilate only the most recent observation, leaving them unable to revise past states when new observations arrive. Estimates then stay tethered to a history that later observations may contradict, and errors accumulate over the assimilation run. To this end, we introduce **DAWIS**, a unified DA method covering filtering, fixed-lag smoothing, and block smoothing within a single framework. DAWIS replaces the single flow time of a state-level prior with a multitask stochastic interpolant over a window of consecutive states, assigning a separate flow time to each. An assimilation cycle inverts the window to a vector of per-state turning points and regenerates it under observation guidance, with the turning points controlling how strongly each state is held fixed, revised, or generated from scratch. The same construction can also absorb the forecast into the assimilation cycle, removing the need for a separate forecasting model. Experiments on challenging nonlinear systems show that DAWIS improves on both filtering and smoothing baselines under sparse, noisy, and nonlinear observations. The code for DAWIS is available at https://github.com/Erik-Wikingsson/DAWIS

Existing machine learning weather forecasting models typically generate forecasts through autoregressive rollouts at a fixed temporal resolution. While highly efficient for long-range prediction, this formulation can suffer from severe error accumulation when used with shorter time steps and does not explicitly encode the locality and temporal continuity of atmospheric dynamics. To address these limitations, we introduce **SDECast**, a Neural Stochastic Differential Equation (SDE) framework for continuous-time probabilistic weather forecasting. SDECast extends SDE Matching to learn stochastic dynamics directly in physical space, without requiring repeated SDE simulation during training. On a simulated geophysical flow, we show that SDECast recovers meaningful drift dynamics and faithfully reproduces the underlying continuous-time behavior. We then demonstrate its scalability to global weather forecasting at hourly resolution, where SDECast produces skillful probabilistic forecasts for lead times of up to five days.

Background: We address the problem of planning when the set of feasible states or actions changes over time. For example, in the problem of path planning among moving obstacles (sometimes known as SIPP), the feasibility of being at a particular location can change as the obstacles move. Or, the action of boarding a particular train is feasible only while it is stopped at the station. This dynamism means that the optimal plan and its duration can change depending on when execution begins. In practice, execution start time is often unknown until planning has completed or another agent gives the go-ahead. However, most prior planning work either ignores dynamism or assumes a known start time. This makes it straightforward to assess state and action feasibility but is impractical for some applications. Objectives: In this paper, we relax the assumption of a known start time. We define the setting of {\em any-start-time planning} and provide algorithms for it. Methods: We present a data structure called a compound arrival time function (cATF) that compactly encodes the optimal plan as a function of start time. We provide general-purpose planning algorithms, based on heuristic graph search, that assemble cATFs by propagating functions along edges instead of scalar costs. Results: We prove that the size of a cATF is at most linear in the problem size. An experimental evaluation of an implementation for the specific problem of SIPP shows that, on difficult problems, agents that rely on replanning often fail, while any-start-time algorithms using cATFs can quickly look up the optimal plan once the execution start time is known. Conclusions: By enabling efficient representations and reasoning for time-dependent plans, this work provides a foundation for planning in dynamic worlds.

Existing convergence analyses of Muon either assume exact orthogonalization or analyze classical Newton--Schulz polynomials, and guarantee only stationarity, so it is unresolved what Muon's five tuned Newton--Schulz steps preserve and whether that suffices to reach a prescribed neural-network training loss. We establish a finite-time training guarantee that accounts for both momentum accumulation before orthogonalization and the tuned finite-step update. For full-batch training of a sufficiently wide two-layer ReLU network with fixed random output weights and a positive-definite limiting neural tangent kernel, we prove that Muon reaches any target empirical squared loss $\varepsilon>0$ with high probability over initialization. For every momentum parameter $μ\in[0,1)$, a target-dependent constant learning rate proportional to $(1-μ)\sqrt{\varepsilon}$ yields a hitting-time bound of $O((1-μ)^{-1}\varepsilon^{-1/2})$, with other problem parameters fixed. The sufficient width is independent of both target accuracy and momentum. The analysis shows that the tuned Newton--Schulz map preserves alignment with the momentum buffer while bounding the update's spectral norm. Control of gradient variation near initialization transfers this alignment to the current gradient, ensuring descent until the target is reached without requiring exact orthogonalization. Numerical experiments support these mechanisms at widths below the sufficient theoretical threshold: gradient-update alignment remains above the analytical reference, and all 30 runs across six widths and five student initializations on a fixed teacher-student dataset reach the target loss while maintaining kernel positivity.

Uncertainty quantification (UQ) for random partial differential equations (PDEs) is ubiquitous in computational science and engineering. However, classical spectral solvers for this class of problems face the curse of dimensionality, and existing neural solvers often ignore the stochastic structure that makes moments and calibration tractable. We introduce a stochastic separable physics-informed neural network, dubbed S$^{2}$-PINN, that represents the solution $u(t,\mathbf{x},\mathbf{Z})$ of a random PDE with a learnable Gaussian spatial dictionary, Fourier temporal features, and a generalized polynomial chaos (gPC) stochastic basis, coupled by a low-rank Canonical Polyadic (CP) tensor decomposition core. The method is trained with a hybrid strong-form and gPC-projected residual loss. Our theoretical analysis establishes that the separable class is dense in $L^2$ under mild conditions, and the projected residual corresponds exactly to a stochastic Galerkin constraint. Furthermore, we show that mini-batch projection coefficients are logarithmically dependent on the number of gPC modes, and that the orthogonality penalty controls the conditioning of the learned spatial dictionary. Using four manufactured random PDE benchmarks, we show that S$^{2}$-PINN outperforms nine baselines in terms of mean and variance accuracy, as well as calibration, while using significantly fewer parameters. Further evaluations on non-manufactured Poisson and Darcy problems, a stochastic Navier--Stokes problem, a diffusion scaling study of higher random dimensions, and two stochastic inverse problems reveal the generalization capabilities of the proposed structure. Together, these results support stochastic separability as an effective design principle for physics-informed neural UQ. The code for the experiments can be found in https://github.com/DMax1314/s2pinn

Complex reasoning queries can be decomposed into directed acyclic task graphs and distributed across heterogeneous LLMs, reducing latency through parallelism and enabling smaller models to solve complex tasks. In practice, however, the suitability of an LLM for a given subtask may be a priori unknown, and execution alone does not reveal output correctness. We propose JOVE, an online framework that jointly assigns executor LLMs and selects intermediate outputs for paid verification. Verification runs asynchronously and is used to improve future allocations, so the system must balance spending on execution now against learning for later. We study how to optimize this trade-off under a long-term budget and a per-query latency constraint, with stochastic, initially unknown LLM service quality, invocation costs, and execution times. JOVE makes execution and verification decisions by solving a sequence of per-query mixed-integer linear programs. Online learning updates task-dependent estimates of LLM quality based on verification feedback, while an information-gain bonus incorporates the value of learning into allocation decisions. Under a natural set of assumptions, we establish sublinear quality-learning regret for JOVE. Across four reasoning benchmarks, JOVE achieves competitive accuracy against standard inference baselines while reducing average cost and latency by at least 3.17 times.

Lung ultrasound (LUS) is attractive for tuberculosis (TB) screening at primary-care level, but labelled cohorts are small. Echocardiography carries no such constraint, while sharing the same underlying ultrasound imaging physics, signal processing and B-mode appearance as LUS. We ask whether an encoder pretrained on that high-resource ultrasound domain carries representations that remain usable in the low-resource one. Only the encoder varies, across seventeen encoders spanning three architecture families. Among them, a latent-predictive video encoder pretrained on generic video (V-JEPA2-L) and its echocardiography counterpart (EchoJEPA-L) differ in pretraining corpus alone. The choice among these encoders does not resolve the classification, the whole family spanning 2.50 percentage points against a measurement resolution of 2.71. What moves the task instead is feature conditioning. Standardising the features between the encoder and the classifier improves all seventeen encoders by a mean of +1.23 percentage points at $p=1.5\times10^{-5}$. On the held-out test set every encoder selected on the development folds stands above the baseline system by up to +2.57 percentage points of area under the receiver operating characteristic curve (AUROC), and specificity at 90% sensitivity reaches 79.3% against 60.3%. The contrast specified in advance, EchoJEPA-L against V-JEPA2-L, measures -0.16 percentage points at $p=0.926$. We therefore find no evidence that shared ultrasonic physics alone makes echocardiography a more productive pretraining corpus than generic video, and any advantage, if present, is smaller than this cohort can resolve. The video encoders receive replicated still images, however, so whether this absence of an effect reflects the pretraining domain or a video encoder applied to static frames cannot be separated. The limiting factor is the labelled cohort rather than the encoder.

Multimodal Large Language Models (MLLMs) achieve strong performance across vision-language tasks, yet the internal mechanisms by which visual and textual information are fused across layers remain insufficiently understood. We investigate representative MLLMs from two architectural paradigms: concatenation architectures and native multimodal architectures. We conduct three progressively connected analyses: alignment decoupling identifies which modality changes, attention routing and entropy characterize how cross-modal information is distributed, and intrinsic dimensionality examines how fusion reshapes feature spaces. Separately, we perform causal intervention experiments as a validation of the resulting interpretation. As a supplementary analysis, we use visual CKA to examine the Platonic Representation Hypothesis. Together, these analyses reveal two distinct fusion pathways: concatenation models follow a text-first, vision-later pathway, whereas native models exhibit earlier visual-textual co-adaptation and feature-space reorganization. This work provides a mechanistic perspective for understanding multimodal fusion and supports architecture-aware diagnostics of multimodal representations.

Reinforcement learning (RL) for learning path recommendation (LPR) faces two coupled obstacles. First, the policy must commit to a sequence of L concepts without intermediate feedback, producing a combinatorial search space that grows super-exponentially with L and provides reward only at the final step. Second, expert learning paths would be the natural cure for sparse-reward RL, but they do not exist in educational data, because student logs record what learners did, not what they should have done. We address both obstacles by importing a recipe from simulator-based demonstration learning in robotics: the knowledge tracing simulator is used both to synthesize per-learner expert demonstrations through evolutionary search and to train a deployment-free policy that distills these demonstrations into a feed-forward learner. Our framework, EVOL, instantiates this pipeline with an asymmetric actor-critic where the actor commits to deployment-realistic blind planning while the critic exploits the privileged simulator state during training. Across three datasets (ASSIST15, Junyi, and EdNet; 39-189 concepts) and path lengths L = 5, 10, and 20, EVOL surpasses 8 baselines spanning heuristic, sequential, RL, graph-enhanced RL, and LLM-enhanced methods. We further compare three imitation strategies (BC, AWR, and DAPG) and show that final performance is governed by the quality of evolutionary experts rather than by the particular imitation objective.

Touché 2026 extends causality extraction to counter-causal claims: news sentences whose surface form appears causal but whose meaning denies the causation, as in "It is falsely believed that X caused Y." A system that relies on surface cues such as "caused" or "led to" will accept such a sentence as causal and give it the wrong polarity. On the Countercausal News Corpus (CCNC), the task has three subtasks: deciding whether a sentence is causal (detection), locating its cause and effect spans (extraction), and labeling its polarity as procausal, counter-causal, or uncausal. We build one model per subtask. Detection is a fine-tuned classifier with a single cross-task rule that uses the extracted spans to remove false positives. For extraction, we ensemble three RoBERTa-large BILOU+CRF taggers by averaging their token-level scores before decoding, rather than voting on the spans each tagger produces. For polarity, where labeled counter-causal examples are scarcest, we add training sentences generated by a large language model prompted with nine patterns of counter-causal expression adapted from Hagen et al., keeping only those that pass automatic structural checks. On the held-out CCNC test set, the system reaches F1 0.869 on detection and macro-F1 0.817 on polarity, and in the organizers' final causal-only evaluation of extraction it scores granularity-adjusted F1 0.728, the highest extraction score among all submissions including the organizers' baseline. The development split is used only for component selection and the ablations reported in the paper.

Large-scale pre-training has improved the generalization of neural operators across diverse PDEs. However, existing PDE foundation models still struggle with heterogeneous dynamics, where shared representations may cause knowledge interference, while mixture-of-experts (MoE) architectures suffer from increasing expert redundancy. We propose SPEAR, a spectral-disentangled MoE neural operator with knowledge-guided expert aggregation for large-scale PDE pre-training. SPEAR decouples latent features into low- and high-frequency components, enabling shared modeling of transferable dynamics and specialized learning of PDE-specific patterns. To address expert redundancy, we design a knowledge-guided expert aggregation strategy that measures expert similarity from dataset-specific learned knowledge and routing preferences, enabling the identification and consolidation of similar experts. Experiments on twelve PDE datasets and multiple downstream benchmarks demonstrate superior performance in pre-training, fine-tuning, and transfer learning. Furthermore, our aggregation strategy reduces the number of experts by 50\% while maintaining or improving prediction accuracy, achieving a balance between model efficiency and generalization for PDE foundation models.

Solving severely ill-posed imaging inverse problems requires recovering image structures that are unobservable or weakly constrained by the measurements. Diffusion models provide expressive learned priors for inferring such missing information, while posterior sampling incorporates measurement consistency along the reverse process. Standard diffusion posterior samplers, however, rely on instantaneous measurement-aware estimates, without explicitly exploiting information carried by previous posterior corrections. We introduce Consecutive Posterior Fusion Denoising Diffusion Null-Space Models (CPF-DDNM), an inference-time strategy that fuses consecutive measurement-aware estimates to improve the diffusive recovery of unobservable image structures, without requiring retraining or additional denoiser evaluations. We instantiate this principle within DDNM, whose range/null-space decomposition reveals that consecutive fusion preserves the measurement-determined component while acting exclusively on the prior-driven null-space estimate. We thus provide a geometric interpretation of CPF-DDNM and a local error analysis that characterizes the optimal time-dependent fusion coefficient, including the extrapolative regime. Experiments on sparse-view and simulated low-dose computed tomography, as well as medical image super-resolution, show consistent improvements over DDNM and competitive performance against diffusion-based inverse solvers.

We release PaMIR (Public Arrival-ordered Measurement for Inference in Risk), an open benchmark for credit-default prediction when labels are scarce and arrive late. The field's reference benchmark studies use eight datasets each, only two or four of them public. PaMIR brings together 19 public datasets with binary default labels -- 1.24M loans, firms and card accounts from nine countries -- rebuilt from pinned source snapshots by one leakage-audited recipe and never redistributed; to our knowledge it is the one of its kind as of today. Every model is a single function, scored under a repeated i.i.d. split and a label-delayed stream in which each application is scored on arrival, with AUC reported by label budget; fleet means are withheld unless every dataset is scored. A synthetic-data harness tests generated training rows without letting a generator see held-out rows. This report describes release 0.4.0 of this living benchmark.

Scalable nonlinear causal discovery requires methods that combine flexible mechanism estimators with efficient search over large graph spaces. Several algorithmic families have been proposed to address this challenge, yet their accuracy-runtime trade-offs remain poorly understood. We empirically compare the four major approaches: differentiable structure learning, amortized structure learning, score-matching, and combinatorial search. Our results reveal complementary bottlenecks: differentiable and amortized methods scale well but exhibit an accuracy gap, score-matching methods can be accurate in low dimensions but degrade quickly for increasing feature sizes, and combinatorial methods remain accurate but are slowed by repeated and redundant local scoring. Motivated by this bottleneck, we develop SPADE, a spline-based score-evaluation scheme that compiles sufficient statistics once and reuses them throughout combinatorial search. Under bounded indegree, its Gaussian variant reduces algorithmic complexity from O(nd^3) to O(nd^2+d^3). Empirically, SPADE shifts the observed scalability-accuracy frontier by orders of magnitude: it solves 100-variable problems with 160K samples in seconds and 1600-variable problems with 2.5K samples in minutes, while retaining high structural accuracy across synthetic and real-world benchmarks. These results reveal a substantial shift in the practical scale of combinatorial search and highlight the importance of evaluating scalable causal-discovery methods along the full accuracy-runtime frontier.

We present a first evaluation of machine learning applied to patient clinical and demographic data gathered in two different countries for the purpose of tuberculosis (TB) screening to identify people who would benefit from expensive molecular testing. Experiments are based on the recently-compiled CAGE-TB dataset, which includes sub-cohorts of people with presumptive TB presenting at community health care centres in South Africa and Uganda. Three neural network architectures (logistic regression (LR), multilayer perceptrons (MLP) and convolutional neural networks (CNN)) are considered in conjunction with greedy feature selection. For the convolutional neural network, a strategy that jointly optimises feature selection and feature ordering is proposed and shown to lead to consistent development and test set improvements. For all three models, development set area under the receiver operating characteristic (AUROC) curve is improved by 2-7% using feature selection. LR after feature selection achieves an AUROC of 0.8 [0.75,0.86] (95% CI) and 0.84 [0.78,0.9] when testing on the held-out Ugandan and South African data respectively. Although outperforming LR on the development cohort, the deeper networks (MLP, CNN) show inconsistent trends on the held-out cohorts, while LR achieves performance within 1-2% of the best achieved in terms of AUROC. LR narrowly misses the WHO minimum requirements by 4-9% in sensitivity even though the network is being evaluated on a completely held-out cohort. The development of neural-network based classifiers for TB screening therefore appears viable.

A model trained on "The capital of X is Y" may produce "Y" after "The capital of X is" but fail after "The capital of X:". We call these different ways of eliciting the same fact request forms. To separate learning a fact from retrieving it, we train two models in two stages. In the first stage (request-form training), one model sees each fact in five forms and the other sees the same facts only as statements. In the second stage (target-fact training), both receive identical training on new facts, all as statements. Both then retrieve the new facts almost equally well from statements, but differ sharply on other request forms. Thus, a model can learn how to retrieve through a request form before it learns the facts. To understand this difference, we examine the hidden state immediately before the answer, which we call the context state. When given two different request forms for the same fact, the model trained on five forms in stage one produces more similar context states than the model trained on statements alone in that stage. Changing this state at retrieval time can enable or prevent retrieval of an already learned fact, and the same effect transfers across facts and factual relations, such as capitals and currencies. To test its role during learning, we change the context state only during target-fact training. This intervention changes later retrieval without intervention at test time. Together, these results show that later retrieval depends on earlier request-form experience and the context state during fact learning.

Large language models (LLMs) are increasingly deployed in public health, finance, and governance, requiring both accuracy and societal value alignment. Despite recent advances, LLMs often perpetuate or amplify bias embedded in their training data, posing challenges to fairness. While self-debiasing encourages an LLM to identify and correct its own biases, relying on a single model's intrinsic knowledge may be insufficient to address deeply ingrained stereotypes. To address this limitation, we introduce Collective Bias Mitigation (CBM), a framework that alleviates bias by learning fine-grained model behavior and fostering knowledge sharing among diverse LLMs. This work is the first to systematically explore the effective selection and organization of distinct LLMs to cultivate fairer LLM responses. Experiments show CBM substantially outperforms standalone baselines (e.g., in the top-7 setting, Committee lowers the age bias score from 0.25 to 0.10). Our Debating and Committee topologies achieve substantial bias reduction, with the latter balancing mitigation effectiveness and inference cost, highlighting the potential of CBM for fairer LLMs.

A natural-language requirement leaves questions open, and a model asked to implement it settles them silently: across 600 generated test suites from three models, 42.7% contain no test that distinguishes the competing readings. An acceptance example written before implementing is the usual remedy, but an example both readings satisfy resolves nothing. We insert two steps into that practice: enumerate the requirement's underspecified points by name, then for each construct two throwaway implementations differing only in that point and keep a candidate input only if executing both shows they disagree. The result is recorded as a decision pin: the named point, the confirmed input, and the value the person chose between the two exhibited results. The same record then constrains generation and decides compliance by execution. On a benchmark of 40 tasks with paired reference implementations and two models, a separating input is obtained for 92.5% of decision points and a pin identifying the intended decision for 85-90%. As checks on 703 independently generated implementations, pins agree with the benchmark's classification on 96-97%, with disagreements on three tasks, one where both classifiers erred. As generation constraints, pins are honoured at the pinned input in all 210 generations and are never worse than a prose rule on held-out inputs in 38 cells, though no better than prose stating the same scope. Every compliance failure under a prose rule came from the model deciding the rule's scope itself; one such case silently overturned another recorded decision, was attributed to a single rule by leave-one-out on both models, and was missed by text-level reconciliation but caught by re-running the recorded input. The setting yields too few such conflicts to evaluate a regression step, and we say why.

We present WAMpy, a Python framework optimized for synthesizing Prolog programs. Unlike general-purpose Prolog systems, WAMpy targets workloads that repeatedly generate and evaluate small candidate programs. WAMpy compiles Prolog clauses into NumPy array-based WAM instructions and supports partial recompilation of hypotheses against fixed background knowledge. Performance-critical routines are accelerated using Numba just-in-time (JIT) compilation. In a benchmark of repeated compilation-and-evaluation workloads, WAMpy improves end-to-end performance compared with SWI-Prolog accessed from Python using Janus.

GPU kernels generated by large language model (LLM) agents can remain less efficient than expert implementations, but runtime alone does not reveal how the gap relates to design discovery and implementation. We introduce D2K-Bench, a diagnostic benchmark of 26 tasks and 85 workloads that measures how effectively agents translate expert design guidance into efficient GPU kernels. The guidance covers L1: high-level algorithmic insights, L2: dataflow design, and L3: low-level optimization tricks, including dependencies among these levels. Pairwise runs with and without guidance share task descriptions, workloads, tools, hardware, and a 350-turn budget. Complementary assessments examine independently proposed designs and the design properties implemented in generated code. Across five models on NVIDIA B200 GPUs, guidance raises correctness over 130 model-task pairs from 93.1% to 98.5% and increases the Performance Score over all 26 tasks from 1.46 to 1.95. For the three frontier models with correct submissions on all 26 tasks in both runs (GPT-6-Astra, Claude-Opus-4.8, and GPT-5.6-Sol), geometric mean speedup increases from $1.69\times$ to $2.49\times$. Across all five models, the mean combined implementation score increases from 57 to 70 out of 100. These results show the value of expert design guidance while identifying design properties that remain unimplemented.

We present the Neuro-Physical Inverter (NPI), a modular, uncertainty-aware framework for geophysical inversion that couples ensemble-based conditioning with constrained residual learning, demonstrated in the 1D magnetotelluric (MT) setting as a controlled testbed. The framework operates in two stages. An Ensemble-Conditional Gaussian Process (EnsCGP) conditions a prior ensemble of resistivity models on the observed response, producing a physically admissible reference ensemble. A residual-learning neural network then predicts targeted corrections to this reference, trained on synthetic data and fine-tuned per station for field application through a physics-coupled objective. Because an ensemble is conditioned, refined, and propagated through both stages, every estimate carries an associated ensemble spread. Synthetic experiments show that NPI systematically reduces ensemble-mean error without destabilizing the ensemble. Applied to broadband MT data from the Gabbs Valley geothermal region (Nevada, USA), NPI reduces the across-station mean misfit over the mid-period band while retaining comparable ensemble spread. The propagated ensemble yields a factor of uncertainty that serves as an operational measure of constraint within the assumed model class. Both stages are dimension-agnostic in formulation, and the design principles established here are intended to scale to higher-dimensional parameterizations.

Diffusion models can synthesise contrast-enhanced CT (CECT) from non-contrast CT (NCCT), avoiding contrast administration and its environmental and patient-access costs. However, visually realistic images are not necessarily anatomically correct, and the pixel-intensity and feature-space similarity metrics used to assess generation quality do not directly measure anatomical correctness. In this work, we investigate whether uncertainty can serve as a proxy for semantic correctness in diffusion-based medical image synthesis. We study NCCT-to-CECT synthesis using AortaDiff, a multitask diffusion framework that jointly generates CECT images and lumen segmentations. The segmentation output provides an explicit representation of the generated vascular anatomy, enabling segmentation-derived errors to be used as a quantitative measure of generation correctness. Six methods spanning weight (Ensemble, HyperDiff, BayesDiff), architecture-perturbation (MCDropout), generative-stochasticity (RDS) and input-perturbation (TTA) uncertainty are compared at the pixel, region and image levels, and for detection of clinically relevant out-of-distribution (OOD) cases. Uncertainty proves informative at all three spatial scales, remains informative on an external multi-centre dataset under distribution shift, and supports OOD detection. MCDropout stands out among the six: it ranks among the leading methods at every scale, generalizes well on the external dataset, and can be enabled at inference on any model already trained with dropout, so reliable uncertainty comes at no extra training cost. Uncertainty reliably flags severe failures but discriminates poorly among already high-quality images. These findings support uncertainty as a practical and computationally economical signal for quality filtering, reliability assessment and OOD detection in NCCT-to CECT synthesis.

Long-horizon LLM agents are typically trained with sparse outcome rewards, making trajectory-level objectives too coarse to distinguish the contribution of individual decisions. Step-level credit assignment provides finer-grained supervision, but its estimates can be unreliable because observed returns also depend on subsequent actions, environment transitions, and trajectory length. We propose AdaStep, an Adaptive Step-credit weighting method that controls how strongly each group-derived local advantage modifies the trajectory-level signal. We formulate this weighting as a mean-squared-error estimation problem for the latent step advantage and, under an explicit conditional sampling assumption, derive an optimal per-state shrinkage coefficient. The coefficient admits a signal-to-total-variance interpretation: it preserves local credit when return variation is attributable to the selected action and suppresses it when variation is dominated by downstream randomness. AdaStep requires only lightweight scalar computation, with no critic, additional rollouts, or extra model inference. Experiments with three model backbones on ALFWorld, WebShop, and ScienceWorld show consistent improvements over baselines at low computational cost.

This paper studies the complexity of convex optimization using lazy second-order oracles (Doikov, Chayti, and Jaggi, ICML 2023), where an algorithm queries gradients every iteration and Hessians once per $m$ iterations. Under this setting, we show a lower bound of $Ω(m+ m^{1/7} ε^{-2/7})$ on the number of total iterations to find an $ε$-solution using a novel block zero-chain construction. Then we propose a novel method that achieves a new upper bound of $\tilde{\mathcal{O}}(m+ m^{1/7} ε^{-2/7})$, which significantly improves the prior one (Chen, Liu, Luo, and Zhang, COLT 2026) of $\tilde{\mathcal{O}}(m+ m^{13/21} ε^{-2/7})$ and is tight up to logarithmic factors.

Hybrid quantum classical neural networks integrate parameterized quantum circuits (PQCs) with established deep learning architectures, but their performance depends strongly on the choice of quantum circuit architecture, a choice that remains largely manual. Most existing approaches rely on hand-designed or fixed circuit ansätze, requiring circuit structure, gate composition, and qubit connectivity to be specified in advance with no guarantee that they suit the task. This limitation is especially acute in image classification, where quantum circuits must transform features extracted by classical networks while remaining compact enough for practical training, requirements that generic, task-agnostic ansätze are unlikely to satisfy simultaneously. We extend EXAQC, an evolutionary framework for automated quantum circuit discovery, to image classification. EXAQC evolves PQCs as intermediate processing modules while retaining classical feature-extraction and prediction layers. On MNIST, Fashion-MNIST, and CIFAR-10, EXAQC achieves 98.42%, 90.62%, and 85.47% accuracy, respectively, while using comparable gate counts to other quantum architecture-search methods. Against classical networks, evolved hybrid models maintain comparable accuracy with substantially fewer trainable parameters, reaching 85.68% on CIFAR-10 with over 25$\times$ fewer parameters than a 10-layer CNN. Encoding choice also matters: rotation-based encodings (RX, RY, U3) outperform amplitude encoding by 22-25 points on CIFAR-10. These results demonstrate that automated circuit discovery yields compact quantum modules that can replace larger classical components in vision architectures while retaining competitive accuracy.

Kernel Singular Value Decomposition (KSVD) learns a pair of singular vectors w.r.t. an asymmetric kernel matrix, which can be induced by two data sources, e.g., the queries and keys in self-attention or the rows and columns of a given matrix. In this work, we extend KSVD to multiple data sources, namely eKSVD, which conducts joint nonlinear feature learning upon asymmetric kernels. In the primal formulation, the projections associated with each data source are jointly learned to capture maximal information, while incorporating pair-wise couplings. With the Lagrangian and its Karush-Kuhn-Tucker (KKT) conditions, the optimization in the dual leads to a generalization of the shifted eigenvalue problem in Lanczos decomposition theorem of KSVD. Further, a covariance-based framework is derived together with using neural networks (NNs) for explicit feature mappings, complementary to the kernel-based interpretation and optimization. Numerical experiments verify the effectiveness of our eKSVD compared to methods based on Mercer kernels for tackling multiple data sources, and our innovation of deploying NNs demonstrates great flexibility for kernel methods.

StanceEval 2026 is the second edition of the StanceEval shared task series on stance detection in Arabic social media text. Stance detection aims to identify a writer's stance toward a given topic. Given a tweet and a target, participating systems must determine whether the writer's stance is Favor, Against, or None. This edition focuses on cross-target generalization across two distinct evaluation tracks: Track 1 evaluates thematically related cross-target transfer (testing on Women Driving, related to Women Empowerment from training data), while Track 2 evaluates cross-domain transfer to completely unseen targets (E-Cars and Trimester System). The shared task attracted 80 registered teams from 12 countries. During the evaluation phase, 30 unique teams submitted entries, with 21 teams officially ranked in Track 1 and 13 in Track 2 following validation filtering, and 20 teams submitting system-description papers. Participating teams employed diverse methodologies, including fine-tuned pretrained language models, prompt-based and retrieval-augmented large language models (LLMs), fine-tuned LLMs, and hybrid cascades. Top systems achieved impressive $F_{avg2}$ scores of 0.8994 on Track 1 and 0.9400 on Track 2, substantially outperforming the strongest baselines (0.7366 and 0.7475, respectively), where $F_{avg2}$ denotes the macro-averaged F1 score over the Favor and Against classes. Counterintuitively, performance on the unseen targets was higher than on the related target, a disparity could be driven by extreme target polarization, class imbalance, and dialectal or sarcastic nuance across topics.

Tool-equipped AI agents use tool calls to access data and act on external systems. Horizontal growth of agentic systems increases the number of these interactions, and further motivates the need for automated, per-call oversight that can operate at low latency and/or on-prem. Conventional authorization schemes can determine whether an agent is allowed to invoke a tool, but cannot assess the agent's underlying cognition, specifically, whether the tool selection represents a logical, relevant step toward satisfying the intent of the task or not. Consequently, an allowed call may still deviate from the task's intent: a rogue agent might deviate the calls or nudge other agents to make a combination of calls that would not align with the intent of the task. Therefore, every call needs to be verified. In this study we investigate the applicability of Small Language Models (SLMs) to this purpose: an SLM functions as a task-tool relevance classifier that evaluates every selected tool independently against the assigned task and returns a relevance signal for downstream enforcement. Equipped with a novel dataset with multi-tool tasks whose required tools span distinct Model Context Protocol (MCP) servers, we used prompt-optimization, supervised fine-tuning, and reinforcement learning through GRPO to optimize and specialize SLMs.

Self-supervised hypergraph representation learning can produce informative node embeddings, but existing methods often require deep encoders trained for hundreds of epochs, making embedding generation costly even for hypergraphs with a few thousand nodes. This limits applications requiring embeddings for many or evolving hypergraphs. We present HyperFuse, a label-free pipeline for fast hypergraph representation learning. HyperFuse (i) computes structural node coordinates by maximizing a spectral relaxation of hypergraph modularity using Banerjee's hypergraph adjacency and a matrix-free operator with cost linear in node-hyperedge incidences; (ii) constructs multi-scale feature summaries and assigns bounded utility weights to hyperedges based on member stability under feature and membership masking; and (iii) trains a lightweight utility-weighted hypergraph encoder for 100 epochs using an invariance-decorrelation objective. We compare HyperFuse with TriCL, SE-HSSL, VilLain, and HypeBoy on nine public hypergraphs using six downstream classifiers and k-means clustering. On the eight datasets where all methods completed, HyperFuse required 8.7 s per dataset on average, achieving 13-179x geometric-mean speed-ups over the baselines. It achieved the highest average accuracy with five of six classifiers, while classification and clustering performance was not significantly different from TriCL and SE-HSSL. Compared with HypeBoy, HyperFuse was 13x faster and 2.1-4.1 percentage points more accurate across all classifiers. HyperFuse provides a practical approach for fast, repeated hypergraph embedding generation.

We establish lower bounds for Hamiltonian property testing with access to the time-evolution operator but not its inverse. Each experiment may query the time-evolution operator multiple times, and distances between Hamiltonians are measured in the normalized Frobenius norm. In this model, we show that testing whether a Hamiltonian is $k$-local or $\varepsilon$-far from every $k$-local Hamiltonian requires $Ω(1/\varepsilon^2)$ total evolution time, matching the upper bound of Kallaugher and Liang (TQC'25). We also prove that testing whether an unknown Hamiltonian equals a target Hamiltonian or is $\varepsilon$-far from it requires $Ω(1/\varepsilon^2)$ total evolution time, matching the upper bound of Sinha and Tong (2025). These are the first lower bounds for natural problems in Hamiltonian learning and testing that rule out Heisenberg-limited scaling of $1/\varepsilon$. As a third result, we show that amplitude estimation to precision $\varepsilon$ requires $Ω(1/\varepsilon^2)$ total time evolution, recovering the result of Tang and Wright (QIP'26) in the continuous-time query model. All three results follow from the hardness of distinguishing the zero Hamiltonian from a suitably chosen ensemble of random Hamiltonians. We establish this hardness by adapting the continuous-time adversary method to forward Hamiltonian evolution.

Flow Matching enables high-quality visual generation via continuous-time dynamics, but inference remains costly due to multiple sequential function evaluations. Existing acceleration methods reduce the number of function evaluations but often introduce additional training overhead, degrade quality, or fail to account for input-dependent variability. We propose COFLOW, an inference-time method that adaptively selects the step counts each generation based on the prompt features. Our context-aware COFLOW is trained online with an unsupervised reward that balances inference efficiency and generation fidelity. Our method is plug-and-play, requiring no retraining of the underlying generative model. It generalizes to image and video generation, achieving over 2.5x speedup while preserving perceptual and semantic quality. We further provide a theoretical analysis establishing an O(1/K) forward-Euler discretization error bound under standard regularity conditions.

Gaussian Processes (GPs) are a powerful tool for modelling and quantifying uncertainty in functional relationships. However, they require practitioners to make a number of design decisions, such as the choice of the kernel and the observation model. Suboptimal choices can produce misspecified models that do not capture the underlying data generating process. We introduce Predictively Oriented Gaussian Processes (PrO-GPs), which treat predictive uncertainty as the primary inferential target and provide a robust alternative to standard GPs. Although direct computation of a PrO posterior for nonparametric models is intractable, we derive a reduced formulation and practical sampling scheme for efficient computation. Through synthetic and real data experiments, we show that PrO-GPs produce better calibrated predictive distributions under model misspecification compared to standard GP approaches.

Large language models fine-tuned from a shared base can be merged by averaging their task vectors, but some merges collapse far below the base model, and common merge operators give no warning before evaluation. We show that one statistic of the specialists' task vectors both predicts this collapse and calibrates its repair. The power that averaging removes equals the variance of the task vectors across specialists, our measure of interference. Under a working noise model, the disturbance that a merge injects grows with the merge coefficient and with interference, yielding a pre-merge score. In our experiments on twenty-two merge configurations from four model families, only destructive merges exceed a threshold on this score. We find that statistics of sign conflict between specialists, a common target of existing merge operators, are anti-predictive. We then predicted the outcomes of fourteen merges before evaluating them, and twelve predictions were correct, including the destructive outcome of a specialist pair pushed past the threshold by continued pretraining. To address this collapse, we introduce PRISM, an operator that averages the task vectors first and then soft-thresholds each layer at a level set by the layer's interference. Without data or tuning, PRISM keeps all five destructive merges above the threshold within evaluation noise of the base model, where plain averaging falls at least 14.4 points below it or collapses entirely. We apply PRISM only above the threshold and keep the plain average for merges below it, which include all fifteen harmless ones. Code is available at https://github.com/js-lee-AI/PRISM.

The memory footprint of the key-value (KV) cache constrains the practical use of long-context models, and it dominates cost when one prefilled context must later serve many different queries. In this reusable setting, query-agnostic compression trades cost against quality: lightweight estimators are cheap but less accurate, whereas full-context reconstruction scoring is more accurate yet reprocesses the entire prompt. We introduce KV$^2$, a query-agnostic KV-cache compression method based on selective reconstruction. KV$^2$ first uses a lightweight proxy scorer to identify informative in-context tokens, then reprocesses only this subset to compute final eviction scores. On RULER, Needle-in-a-Haystack, and LongBench, KV$^2$'s margin over baselines widens as the budget tightens: on RULER 16K at a 2% KV-cache budget it improves the average score over the next-best baseline by more than 40 percentage points, and on LongBench it attains the highest average across 2%-10% budgets at lower compression-stage runtime and peak memory than full-context reconstruction. Reusable KV-cache compression thus does not require reprocessing the full context. Our code is available at https://anonymous.4open.science/r/KVsquared-0B97.

As LLM agents decide on users' behalf which product to buy, which hotel to book, or which paper to cite, a preference for items from certain sources (the sites or services they come from) shapes what users receive and which sources are selected. We study source preference in end-to-end search with 12 agent models across three domains. Comparing items from different sources that satisfy the same requirements at the same position, we find that each model prefers some sources and avoids others in every domain, largely agreeing on which. This preference can outweigh how well items satisfy the request: an item satisfying one requirement fewer is selected about two-thirds of the time when it comes from a preferred source and the better one from a dispreferred source, but almost never in the reverse case. The information identifying an item's source affects selection by itself: hiding it weakens the preference, and relabeling an item with a preferred source raises its selection rate. We test two routes to this preference: training that rewards better items can make a source a shortcut for requirement satisfaction, and missing information can trigger preconceptions about the source. Supplying missing information or a prompt countering these preconceptions reduces source preference.

Accident, defect and outage investigations end with a decision that ordinary question answering never faces: whether the evidence gathered so far is enough to close the case. We study this decision for LLM investigators, which request evidence from a case file, revise their hypotheses, and either close the case with a conclusion grounded in what they read or leave it open and name what is missing. This judgment does not come with capability: an untrained 9B model overstates its evidence in 97% of its answers, and a frontier model that identifies the right cause in 84% of cases still overstates in 91% and closes 17 of the 41 cases whose official finding is "cause undetermined". Measuring it is also non-trivial: the source of a case largely predicts its label, and a rule that reads only the source reaches 83.0 balanced accuracy on our test cases. We therefore evaluate closure with three tests: closure accuracy, reported against this rule and within each source; evidence dependence, which removes the grounds of a conclusion and checks whether the model stops closing; and conclusion and gap quality, a judged checklist of what the model asserts and what it says is missing. We build Nautil, 731 audited cases from aviation, rail, maritime, chemical-safety and vehicle-defect reports and production server incidents, with teacher trajectories, an out-of-distribution test set and counterfactual evidence versions. Fine-tuning a 9B model on these trajectories makes its closures follow the evidence: removing the grounds lowers its closure rate by 26 points relative to a matched control, overstatement falls from 97% to 35%, and correct, non-overstated conclusions rise from 3% to 43%. Reinforcement learning that rewards only the closure decision then raises balanced accuracy from 69.2 to 83.3, on par with the teacher, and within-source accuracy from 60.4 to 74.1, at some cost in evidence dependence.

On-policy distillation (OPD) has become an important approach to language model post-training. However, despite its performance gains, OPD can also collapse into excessively long and repetitive generation, and the mechanism underlying these divergent outcomes remains poorly understood. We explain these outcomes through a reinforcement learning perspective: the teacher implicitly rewards student behaviors, even those it rarely exhibits itself. From this perspective, our experiments show that OPD improves performance without expanding the student's capabilities. When the implicit reward model is reliable, OPD makes correct responses easier to sample. In contrast, when the preference misaligns with quality, reward hacking happens: the implicit reward model amplifies overlong, repetitive student rollouts, even though it rarely generates such text itself. Guided by this diagnosis, we find that masking unhealthy responses during training and using SFT initialization can each effectively mitigate the collapse. Together, these findings show that OPD amplifies student behaviors favored by the teacher's implicit feedback, shifting the focus from how well the teacher generates to how reliably it evaluates student rollouts. Our code is available at https://github.com/HancCui/opd_hacking.

We study hindsight-guided distillation for rare disease diagnosis on ZebraMap: a 1.5B student is fine-tuned on chain-of-thought traces from a 8B teacher that observes the ground-truth diagnosis during generation. Absolute accuracy remains low for all models - the task is hard at this scale - but within this ceiling a filtered variant (StudentF) achieves a small, statistically significant accuracy advantage over the teacher (p < 0.001), concentrated in better-represented diseases. The unfiltered student does not significantly outperform the teacher (p = 0.129), establishing that contamination filtering - not hindsight distillation alone - drives the gain. The gap traces to an artifact we term GT hallucination. Label-visible generation causes the teacher to embed "ground truth is X" phrases in its reasoning chain; SFT copies the pattern. At inference, the unfiltered student reproduces the phrase in 33.9% of cases, with severe accuracy degradation when the hallucinated label is wrong. A regex filter removing these slots reduces contamination to near-zero, producing the observed gain - though the effect remains small. We precisely quantify this gain-cost tradeoff, document frequency-dependent knowledge transfer absent from the RL-trained teacher, and characterize a calibration gap that SFT does not close - identifying both as directions for future work.

Digital watermarking supports source attribution for AI-generated images, but its reliability depends on resistance to removal attacks. Some attacks attempt to remove watermarks by forcing the decoded watermark to differ from the original. However, this can produce an inverted watermark that remains detectable, causing removal to fail, while further attempts to alter the watermark may unnecessarily degrade image quality. To address these limitations, we present LiBRA (Latent In-band Bidirectional Removal Attack), which aims to make watermarks undetectable while preserving image quality. Instead of continually pushing the watermark toward inversion, LiBRA adjusts the image to conceal the watermark without encouraging further changes that could degrade image quality. Some attacks keep pushing decoded bits away from the original watermark, even when further changes preserve detectability and damage image quality. With access to the watermark key and decoder, LiBRA makes bounded changes in a public autoencoder's latent space. Unlike inversion-driven objectives that cannot correct excessive inversion, LiBRA guides average decoding confidence toward random guessing from either direction. This helps avoid an inverted but detectable watermark. Leaving individual bits flexible allows image-quality constraints to favor less damaging changes, while an optional frequency-guided mask limits their location. We verify removal using an exact two-sided binomial test rather than assuming the confidence target guarantees success.

Several variance-reduced versions of REINFORCE based on importance sampling achieve an improved $O(ε^{-3})$ sample complexity to find an $ε$-stationary point, under an unrealistic assumption on the variance of the importance weights. In this paper, we propose the \algo (Defensive Policy Gradient) algorithm, based on defensive importance sampling, which achieves the same rate without any assumption on the variance of ordinary importance weights. We also establish lower bounds in a generalized black-box policy-optimization model that hides states and actions and permits parameter-dependent rewards. In this model, the optimal rates are $Θ(ε^{-4})$ with bounded-variance one-policy feedback and $Θ(ε^{-3})$ with mean-square-smooth coupled two-policy feedback. Under standard policy-regularity conditions, REINFORCE and \algo realize the corresponding oracle conditions and attain the $O(ε^{-4})$ and $O(ε^{-3})$ upper bounds, respectively. Although the lower bounds do not apply directly to the classical MDP interaction model in which these algorithms operate, this correspondence provides oracle-level evidence that the faster rate of \algo is optimal and genuinely separated from that of vanilla policy gradient.

Adapting large language models to an individual author's style from a few examples is challenging, and scientific writing sharpens the difficulty: formal conventions leave little surface variation, and authors write about their own topics, so extracted ``style'' easily entangles with content. We study style-conditioned abstract generation from a few example abstracts per author and propose three methods: (1) contrastive activation steering, (2) a network that predicts steering vectors, and (3) a hypernetwork that predicts LoRA adapters. We find a consistent trade-off between style imitation and output quality: fine-tuning buys most of the available style signal but forfeits fluency, while the hypernetwork achieves the best trade-off on both seen and unseen authors. Our steering operates at author level, contrasting an author's abstracts against style-neutral generations for the same content. This holds topic fixed, removes the need for a predefined style inventory, and outperforms inventory-based steering. % [EDIT 1a] softened "no single optimal axis" claim Moreover, our analyses demonstrate that manually extracted and predicted steering vectors are near-orthogonal yet score comparably, indicating that style conditioning here can admit at least two unrelated directions rather than requiring one particular axis.

Feature selection for imbalanced classification tasks such as credit card fraud and consumer default detection requires balancing predictive relevance, inter-feature redundancy, and computational feasibility. We benchmark three computing paradigms, classical branch-and-bound optimization (Gurobi), photonic entropy computing (QCI Dirac-3), and simulated photonic boson sampling (Piquasso), across thirteen feature-selection methods on two datasets: ULB Credit Card Fraud (30 features) and AmEx consumer default (159 features). Each method is routed to the solver matched to its mathematical structure. On ULB, Dirac-3 MI-Spearman matches the all-features model using 13 of 30 features (mean F1 0.873 +/- 0.023 over five runs, best run 0.896), and Piquasso is the best method at k=5. On AmEx, performance rises steadily with the feature budget and every paradigm approaches F1 = 0.80 only near the full feature set. Most differences between Gurobi and Dirac-3 on identical methods fall within run-to-run variation; the large gaps occur where the certified optimum generalizes poorly, most sharply for distance correlation on AmEx at k=25 (Gurobi F1 = 0.422 vs. a Dirac-3 mean of 0.746). At matched budgets, F1 varies about ten times more across methods on ULB than on AmEx, which we trace to how concentrated the predictive signal is in each feature space.

Discovering broadly neutralizing antibodies (bnAbs) from human natural immune repertoires remains a fundamental challenge in immunology, hindered by: the extreme rarity of bnAb, incomplete understanding of their cellular origins across pathogens, and the inability of existing computational tools to generalize across emerging viral threats. Here we present ImmuneAgent, a closed-loop AI system that integrates multimodal reasoning with continual meta-learning and wet-lab feedback to overcome these barriers. Applied to screen the natural BCR repertoires from vaccinated or infected cohorts, the system achieves a ~55% neutralization antibody discovery rate (60 of 110 cloned candidates) and a ~11% bnAb yield (12 of 110), substantially outperforming a state-of-the-art sequence-based neutralization predictor or cofolding models evaluated at the same cloning budget. Five ImmuneAgent-discovered antibodies conferred 100% in vivo protection against lethal influenza challenge, comparable to the clinical-stage therapeutic MEDI8852. The system recovered the cellular and structural determinants of bnAb activity and identified FCRL5+CD27+ atypical memory B cells as a conserved bnAb reservoir and hydrophobic interface enrichment as a cross-viral structural signature, which generalized to unseen antigens, discovering human metapneumovirus (hMPV) cross-neutralizing and human papillomavirus (HPV)-neutralizing antibodies without antigen-specific sorting. These results validate that ImmuneAgent is a generalizable framework for rapid therapeutic antibody discovery against emerging viral threats.

Video generation models produce strikingly realistic sequences and are increasingly proposed as world models, yet recent benchmarks reveal pronounced deficits in their physical reasoning. This raises the question of whether these models internalize physical principles or merely reproduce familiar motion patterns. We address this by probing internal representations of video Diffusion Transformers (DiTs) for simulator-derived ground-truth physical quantities spanning kinematic motion and rigid-body dynamics under gravity and contact. We find that these quantities are linearly decodable with high accuracy early in the denoising process, substantially outperforming a baseline decoded directly from the model's own noised latents, indicating that the relevant physical information is actively constructed during denoising rather than already present in the input. Additionally, we show that activations at on-object tokens carry the relevant physical information and that quantities defined over multiple frames are readable from single latent frames. Hence, information is sharply localized within the token sequence and is computed globally but stored locally. The probes further show partial extrapolation, transferring to scene variations and object configurations outside their training regime, so what they read is not simply a correlate of the scenes they were fit on. When fitted directly in the full-resolution activation space, the probing directions can serve as steering vectors to change the model's output.

Workspace agents combine large language models with execution harnesses to perform stateful, multi-step tasks that access or modify external resources. Existing benchmarks leave gaps in executable coverage of their runtime security risks, while evolving model capabilities, harnesses, tools, and threats motivate benchmark evolution. We introduce EvoRiskBench, an evolving benchmark organized around the EP-Path-EF framework, which links an initial risk entry point to a one-hop technical effect through an agent-mediated risk path. The framework defines nine entry-point categories and five effect categories; a 20-participant study supports their interpretability and classification consistency on representative cases. Guided by this framework, an automated end-to-end workflow constructs and executes risk cases in isolated environments and independently verifies outcomes using runtime traces and environment states. The benchmark provides a reproducible dataset of 450 adversarial tasks across six scenarios. We evaluate nine model-harness configurations spanning three models (GPT-5.6 Sol, DeepSeek-V4-Pro-0813, and Claude Opus 5) and three harnesses (Claude Code, Codex, and OpenClaw). Our results reveal substantial vulnerabilities across systems. The most vulnerable configuration, Codex with DeepSeek-V4-Pro-0813, reaches a 68.44% attack success rate (ASR), indicating that configuration of workspace agent is insufficient to ensure secure autonomous execution. ASR varies more across models than harnesses, and harness differences depend on the model. The benchmark cases and evaluation platform will be released after completion of artifact safety and reproducibility checks.

Streaming sensor applications routinely suffer from delayed or missing observations caused by faults, communication losses, or environmental interference. Although recent imputation methods exploit temporal and spatial dependencies effectively, most either assume offline access to future observations or prioritize throughput without enforcing domain plausibility. We present TSGuard, a real-time demonstration system for monitoring, validating, and imputing missing values in streaming time series. TSGuard combines a lightweight graph-aware temporal imputation model with constraint-aware validation, fallback estimation, and operator-facing explanations. Rather than treating imputation as an isolated prediction task, TSGuard integrates it into a broader data-quality loop: detect problematic observations, impute missing values, validate estimated against physical and spatial constraints, and either retain the original value as a plausible anomaly or replace it when it violates domain constraints. Using environmental sensing as a motivating setting, the demo enables users to inspect delayed sensors, compare imputers, define constraints, and validate flagged values in real time. The combination of lightweight online spatiotemporal imputation, domain-aware validation, and explicit retain-or-replace decisions is our central contribution, while interactive explanations make these decisions inspectable and actionable. for operators.

Specifying a goal in language rather than as a goal frame is a natural interface for planning with a latent world model, but testing it needs scenes in which language must discriminate between several objects. We build SLIM, a pushing benchmark with several small objects and paired visual and language goals on identical scenes. On SLIM a LeWM world model that solves PushT succeeds on under 1% of trials, although a scripted controller with simulator state solves every tier. Probes locate the failure in the encoder: its latent is nearly action-insensitive, neither pusher nor object positions can be decoded from it, and rollouts are no better than copying the current latent forward. One inverse-dynamics auxiliary loss, applied to encoder latents and to predicted latents through a shared head discarded at test time, restores every probe and raises success from 0.003 to 0.35 (0.16 on the hard pushing tier, where a goal-agnostic policy scores zero), and improves PushT at twice the trained horizon. Controls attribute the repair to the gradient into the encoder, and a response sweep shows that the vanilla model plans once enough of the frame responds to actions. A cheap action-sensitivity probe, computable without environment access, acts as an empirical necessary condition: all configurations below its threshold failed to plan. On the repaired latent, a small language-goal head plans from sentences without retraining the world model: it reaches 0.84 on navigation (visual-goal oracle 1.00), follows the named zone when it is swapped with a decoy, and degrades gracefully to unseen nouns. A single goal sentence rarely completes a push, but given the push as a sequence of stage sentences the head raises success on the medium and hard pushing tiers from 0.04 to 0.25, on par with the goal-frame oracle, also when the switch between stages is read from the latent alone.

Reasoning traces improve large language models (LLMs), but current models are trained to reason mostly in English. It has been shown that forcing a model to reason in another language degrades accuracy, even when the reasoning language matches the language of the prompt -- but only for a setting where the model reasons over a short prompt. Here, we ask whether the same holds for retrieval-augmented generation (RAG), where the model must read and integrate a large amount of retrieved evidence in the target language. To study this, we build a fully monolingual German RAG question-answering testbed over the fictional world of the tabletop role-playing game The Dark Eye, a domain that is richly documented in German but too niche for the model to answer from memory, so that it has to rely on retrieval. Varying the forced reasoning language of an agentic RAG system on this testbed, we find that aligning the reasoning language with the language of the query and the retrieved documents helps. Forced German reasoning outperforms forced French, although the model benchmarks higher in French, so the benefit comes from alignment and not from language proficiency. The advantage grows when the retrieved context is richer and structure-aware. However, forced German only reaches the level of the model's native, unconstrained English reasoning without surpassing it, showing that native multilingual reasoning is needed. We publicly release the testbed and QA benchmark.

Serving long-context autoregressive language models is constrained by the key-value (KV) cache. Most dynamic eviction methods score token importance per query head and choose tokens independently. This fits poorly with grouped-query attention (GQA), where several query heads share one physical KV buffer: divergent per-head selections force the serving engine to retain the union of their choices - inflating the cache by up to the group ratio r - while arithmetic mean pooling dilutes the specialized retrieval heads that carry factual recall. We introduce Page-EntroKV, a formal framework for KV-cache eviction operating at the granularity GQA serving actually allocates. Heads within each physical group are pooled by weights derived from sink-isolated collision (Renyi-2) entropy - one inner product per head, computed once at prefill with no calibration - so sink heads cannot masquerade as retrieval heads. Pooled scores are projected onto PagedAttention page frames, and eviction executes at the hardware tuple (layer, group, page). We formalize the union overhead ratio (UOR) and intra-group disagreement, prove an exact identity linking them for two-head groups alongside two-sided bounds at every group ratio, prove strict budget preservation and a finite-context needle-retention bound that arithmetic mean pooling provably violates, and give exact per-layer page accounting. On a pilot architecture (Qwen2.5-1.5B-Instruct, r=6), head-independent replay over 2,240 group measurements yields union overhead up to 4.75x at a 2% budget, while Page-EntroKV holds UOR exactly 1.000; sink isolation removes a 13x sink masquerade; needle recall is 100% versus 0% for mean pooling at a 20% budget; retained cardinality is exact for every page size; and QA and code tasks remain solvable at 20% retention.

Generative planners based on diffusion/flow matching can learn to synthesize long-horizon trajectories from demonstrations. However, real-world deployment requires (i) enforcing safety constraints during execution and (ii) tight online replanning at fast execution rates. Prior safe diffusion/flow planners generate the agent's full trajectory at once, while repeatedly perturbing intermediate states to satisfy safety constraints. This approach is not only computationally intensive, but also introduces distribution shift since the learned sampling dynamics is distinct from the system's execution dynamics. We propose SafeStreamingFlow, a goal-conditioned planner that aligns flow sampling dynamics with execution dynamics by sequentially integrating a learned state vector field with hierarchical state prediction. Importantly, we need to enforce safety constraints only for the executed step via high order control barrier functions. Across navigation, racing, and locomotion benchmarks, SafeStreamingFlow reduces planning latency and improves safety compared to existing methods, while maintaining competitive goal-reaching success.

Biological literature retrieval systems are often developed and evaluated using broad biomedical corpora and general-purpose search tasks. However, many curated knowledge bases operate in narrower model-organism domains, where the literature is sparse and terminology is organism-specific. We introduce a retrieval benchmark from dictyBase for Dictyostelium, a model organism in cell and developmental biology. The benchmark consists of curator-generated biological queries linked to PubMed-indexed articles, together with structured gene annotations. Using this benchmark, we study three factors in niche biological retrieval: cross-encoder reranking, gene-aware query expansion, and abstract-only versus full-text retrieval. We report that reranking and gene-aware query expansion improve retrieval selectively: reranking is most useful when the model is well suited to biological evidence matching, whereas curated annotations help clarify compact biological queries by reducing vocabulary mismatch. Full-text chunks substantially improve retrieval when abstracts omit supporting evidence, increasing both candidate recall and top-rank performance, although these cases are harder than queries supported by abstracts. Data and code are publicly available at https://github.com/fulaibaowang/dictycite, and the benchmark dataset is additionally archived on Zenodo.

Quantitative trading strategy design aims to discover trading programs from historical data that remain effective in future markets, which can be viewed as a black-box program optimization problem. LLM-based textual gradients offer a promising approach by providing explicit optimization directions for iterative strategy refinement. However, directly applying textual gradients faces two challenges: (1) optimization is myopic, underutilizing experience from previous evaluations; and (2) aggregate backtest feedback overlooks temporal robustness, potentially favoring strategies that perform well only in specific market periods. To address these challenges, we propose TradeGrad, an experience-guided textual-gradient framework for robust trading strategy optimization. TradeGrad leverages accumulated optimization experience to estimate textual gradients and employs multi-scale revisions for both strategy exploration and refinement. It further introduces the Cross-Period Robust Objective (CPRO), which emphasizes performance in unfavorable historical periods to promote temporal robustness. Experiments on cross-sectional and time-series strategy design in Chinese A-share and U.S. equity markets show that TradeGrad achieves the best in-sample and out-of-sample performance across all four settings. Notably, its Chinese cross-sectional strategy achieves 27.99% annualized return, 12.19% maximum drawdown, and a Sharpe ratio of 1.63, approximately 68% higher than the CSI 300 benchmark. Further analyses validate the proposed components and show consistent improvements in both in-sample and out-of-sample performance throughout optimization. The code is available at https://github.com/transcend-0/TradeGrad.

Hardware-aware neural architecture search (NAS) is dominated by evaluation cost: every architecture must be trained before its reward is known. Conformal-prediction filters cut this cost by pruning candidates whose predicted-reward upper bound misses a threshold, with a distribution-free guarantee that at most a fraction $δ$ are wrongly discarded. That guarantee assumes exchangeability between calibration and test candidates, which the surrounding reinforcement-learning (RL) loop violates: the policy's proposals improve as search proceeds and, in layer-by-layer construction, shift within every episode. We replace one-shot quantile estimation with online feedback control (Adaptive Conformal Inference, with tuning-free, locally-adaptive, and group-conditional variants), restoring steerable coverage: dialing the target delivers it, monotonically and reproducibly, for arbitrary sequences. Across three neural-network architecture families and both single-step and sequential search (three seeds), it tracks every requested level to within ${\sim}10^{-3}$ while pruning 25-50% of evaluations at no measured accuracy cost, whereas static calibration loses control of its coverage and a Gaussian-process baseline stays conservative regardless of the request. Finally, used as an acquisition function on one constrained testbed, the same optimistic bound beats random search, a gain that fixed optimism already carries and online calibration sharpens. The source code is available at https://github.com/Vicomtech/rl-hw-nas.

Speech delivery varies with both the requested paralinguistic attribute and the linguistic content. We introduce ParaGeo, a matched-content decomposition of paralinguistic variation in a frozen speech language model. Synthesized audio tokens are replayed with a fixed listening prompt; pooled key/value (K/V) representations are centered and projected into a shared low-dimensional space. Our GLM-4-Voice probe spans 80 requested controls from 12 benchmark families across eight sentences. With a globally fitted calibration basis, content-held-out centroid accuracy using this basis is 9.49% versus a 1.25% permutation baseline; same-label cross-content cosine similarity is 0.285 versus 0.017, and both conditional permutation tests yield p = 0.001. A separate ten-scenario, six-style probe reveals reproducible contrast directions across scenarios. Static, additive, and temporal interventions produce attribute-, layer-, and schedule-dependent response profiles. These results provide a shared coordinate representation for measuring paralinguistic structure and an empirical starting point for latent speech control. Code is available at https://github.com/yuhanlydia/ParaGeo.

Open-weight language models can be downloaded, modified, and deployed beyond their developers' control, limiting the effectiveness of centrally enforced safeguards. Recent work has therefore proposed \emph{trigger-tag} mechanisms that produce a detectable signal when a model is used under a target condition, such as generating phishing contents. Although these mechanisms borrow from established techniques, their use for conditional misuse detection in open-weight LLMs is relatively new. Therefore, existing research works have not systematically studied the robustness of trigger-tag mechanisms under adversarial attacks. To close this gap, (i)~we formalize trigger-tags and distinguish \emph{token-level trigger-tags}, which introduce watermark-inspired signals during decoding, from \emph{weight-level trigger-tags}, which learn backdoor-inspired associations between target conditions and detectable model behavior. Furthermore, (ii)~we introduce \Untag, a unified attack framework that organizes their mechanism-specific attack surfaces into a common taxonomy. We evaluate representative token-level and weight-level trigger-tags using phishing as a case study. We find that while trigger-tags may provide useful evidence in controlled settings, our attacks render the existing trigger-tag mechanisms to be entirely ineffective. Consequently, we argue that these mechanisms should not be treated as robust misuse detectors when attackers can transform outputs or modify open weights.

Existing streaming vision-language models (VLMs) continuously perceive and reason over visual streams, but their computational pathways remain fixed throughout inference. Consequently, they cannot adapt computation to evolving scene dynamics, where different future events demand different levels and forms of perception. We show that streaming VLMs inherently possess the ability to anticipate the immediate future, and leverage this capability to dynamically configure future computation in a training-free manner. Realizing such anticipatory computation, however, is very challenging: future anticipation must be sufficiently reliable to guide computation, planning must run concurrently with streaming inference, and online reconfiguration must incur negligible overhead. To address these challenges, we introduce FORESIGHT, a dual-stream architecture comprising two Siamese LLMs with shared weights, input encoders, and KV cache. The first LLM continuously processes incoming tokens, while the second runs ahead of the stream to anticipate future context, plan future computation, and generate task responses without interrupting streaming inference. Each plan decides when to reason next, what to check then, and how densely to sample, keeping transient evidence separate from persistent control. The resulting computation plan is executed online through an efficient reconfiguration protocol with schemaguided decoding and lightweight diff-based updates, enabling dynamic adaptation with low overhead. With a frozen Qwen3-VL-8B backbone, FORESIGHT achieves 23.0 mean joint F1 on OmniPro Online evaluation beating strongest trained baseline by 9.5%, while improving the backbone by 6.7 on StreamingBench and 15.4 on OVO-Bench, with the largest gain of 18.7 when evidence arrives later in the video stream. Our source code will be made publicly available.

In lifelong reinforcement learning, retaining previously learned policies is not sufficient for effective transfer to a new task. Useful knowledge may be distributed across several prior policies, and its relevance may change as the learner acquires experience. One hypothesis is that task similarity can be effectively used in a continual learning setting to find and combine previously learned policies. To test it, Adaptive Mask Selection and Composition (AMSC) is designed to estimate similarity from online experience via non-parametric Wasserstein task embeddings from state-action-reward samples. The z-score-normalized sparsemax of the similarity scores are used to derive a variable-size support to periodically choose and weight policies to form a prior when learning a new task. On CT-graph and MiniGrid, AMSC achieves higher mean performance and forward transfer than the evaluated modular composition baselines while exhibiting no forgetting. Results on Continual World suggest that identifying relevant prior knowledge and determining its layer-specific composition may require additional layer-specific tuning. Ablations show that selecting relevant sources and determining how strongly to reuse them are central to these gains. Independently measured pairwise transfer is also positively associated with task-embedding similarity. These results indicate that task similarity can be an effective criterion to select and weight specific knowledge for reuse in lifelong reinforcement learning.

Deep Neural Networks (DNNs) inherently exhibit a degree of robustness to bit-level faults due to their distributed representation of information. As a model increases in width, this information becomes more dispersed, theoretically reducing the impact of any single bit fault. In this paper, we empirically investigate the relationship between model width and robustness to Single Event Upsets (SEUs). We conduct a comprehensive experiment in which baseline models undergo iterative structured pruning to reduce their width while preserving task performance as much as possible. At each pruning stage, we run a targeted fault-injection campaign to evaluate the model's performance under simulated bit-flip scenarios. Our results show that, although structured pruning increases per-inference sensitivity to faults by reducing redundancy, this effect is effectively counterbalanced by shorter execution time, which lowers the probability of encountering an SEU. These findings suggest that structured pruning can yield significant energy and latency savings without compromising overall reliability, providing useful guidance for designing robust AI systems for space applications.

Matching candidates to vacancies is central to recruitment, and a recruiter needs to see why a candidate fits, not only a single opaque relevance score. We provide this evidence as named, interpretable matching dimensions recruiters can act on - eight in our current deployment. We propose a two-part approach. The first is an LLM-based labeler whose prompts and feature definitions were refined from recruiter feedback while it served as an earlier production matching stage. In the current architecture, it is used only for offline labeling and is not called on online requests. The second is a feature bi-encoder distilled from it: a LoRA-adapted embedding backbone with compact per-dimension heads that runs on CPU and serves all online requests. Both parts keep improving: prompts are revised as feedback arrives, and the bi-encoder is retrained on the updated labels. The model is trained on 168,772 labeled vacancy-resume pairs (17,921 vacancies and 180,030 resumes). Recruiters using the service can confirm or revise surfaced feature predictions. On 927 recruiter-recorded values from this selected production-feedback subset, the deployed student agrees with the recorded decisions in 888 cases (95.79%). This is operational, non-blinded agreement rather than an independent human evaluation.

Supervised AI-text detectors report high benchmark accuracy, but it is not clear what their decisions are based on. We analyze a RoBERTa-based detector under semantic, structural, and tokenizer-level perturbations, using the M4 dataset (N = 10,000) and controlled generations (N = 300). When Mistral-7B-Instruct was asked to make machine text sound more human, Verb Diversity rose from 0.77 to 0.92 and the outputs became easier to detect. Detection scores appear to track statistical complexity, which also leads to a 76.3% false-positive rate on formal human writing. As a control, we evaluate event-based Latent Space detection. Paraphrasing changed 87% of its event sequences (Jaccard = 0.067), and homoglyphs altered 70% of the extracted verbs even though extraction still ran (Jaccard = 0.30). Its best domain AUC was 0.577. RoBERTa's robustness seems specific to the features it uses, and structural abstraction did not make detection more robust.

While representation similarity across independently trained language models is well-documented, how internal mechanics such as attention behave across models remains far less characterized. Inspired by this gap, we examine the structure of post-softmax attention weights by marginalizing over query positions, mapping them into a joint token-head "marginal attention space". Evaluating across 60+ diverse LLMs, we find that different properties emerge when reducing this space along its token and head axes. When reduced token-wise, marginal attention yields a text-intrinsic signal robustly conserved across models. To explain this property, we empirically connect marginal attention to the input-output Jacobian of the network, and prove theoretically that under a smoothness assumption, models with similar next-token distributions are guaranteed to have similar input-output Jacobian statistics. When reduced head-wise, it forms a model-private signature conserved across documents. Practically, this provides a natural way to estimate a per-head budget for key-value (KV) cache eviction, effectively decoupling model-specific budget allocation from text-intrinsic token scoring. On standard eviction benchmarks, a per-head budget precomputed offline on pretraining text, combined with a training-free token score, shows competitive performance with methods that recompute the budget on every document or train it per target. Code available at https://github.com/Flegyas/marginal-attention

The subseasonal-to-seasonal (S2S) timescale, roughly from two weeks to two months ahead, is a critical forecast window for sectors such as agriculture, energy, and water management. Yet, it is widely known as the `predictability desert'. Recent AI weather models excel up to two weeks ahead but deteriorate beyond, largely because they are trained to predict fine-scale details that are neither predictable nor essential at S2S timescales. We argue that a more physically grounded objective is to forecast only the slowly varying components that remain predictable. Computer vision reached the same conclusion with the Joint-Embedding Predictive Architecture (JEPA), which predicts in latent space, discarding unpredictable details. In this work, we introduce S2S-JEPA, which brings the JEPA paradigm to S2S forecasting. It is tailored to this task through design elements from state-of-the-art AI weather models. S2S-JEPA achieves comparable skill to the gold-standard ECMWF physics-based ensemble and surpasses it on multiple metrics at weeks 5 to 6.

An LLM agent can recognize uncertainty yet still choose the wrong next step: asking when action is already justified, or seeking clarification when the constraints must change. We formalize actionable indeterminacy: act when an accepted action is shared across all admissible preferences or objectives, clarify when each possibility is feasible but no action is shared, and propose a minimum-cost permitted constraint repair when the request is infeasible. We construct a solver-grounded benchmark spanning object allocation, meeting scheduling, apartment choice, and stable matching. Matched pairs retain the same source while changing whether intervention is necessary, and evaluation separates decision correctness, matched-pair reliability, and fully correct responses. Our findings reveal a recurring difficulty in recognizing when intervention is unnecessary: models can identify situations requiring clarification or repair yet still intervene when a justified action already exists. Correct decision labels also fail to guarantee usable actions, questions, or repairs. Crucially, response requirements shape not only how decisions are expressed but also which decisions are made. Making the required content explicit substantially improves fully correct responses and can change intervention decisions, even when outputs are already parseable. These findings highlight that reliable agency requires more than recognizing uncertainty: it requires intervening only when necessary and translating the chosen next step into a verifiable response.

Clustering variables in causal graphs reduces the size of the graph and simplifies causal inference. However, arbitrary clustering can alter crucial causal relations among variables and lead to erroneous conclusions. While the identifiability of a causal effect in the clustered graph implies the identifiability in the original graph under mild conditions, nonidentifiability in clustered graph does not imply nonidentifiability in the original graph without further assumptions. When both identifiability and nonidentifiability are preserved, the clustering operation is called identification invariant. We present a broad class of clustering operations that are identification invariant based on conditions related to the c-components of the original graph. Finally, we demonstrate use of the results in practical settings.

E-commerce videos are information-dense and frequently compared by consumers evaluating products and merchants assessing marketing strategies. However, existing multimodal models mainly focus on single-video understanding and have limited ability to compare information across videos. We introduce AdsCVR, the first e-commerce cross-video reasoning benchmark, containing 2,483 videos and 6,110 question-answer pairs across six reasoning dimensions. Cross- video reasoning requires models to locate fine-grained evidence among many redundant frames and integrate visual details, speech, and on-screen text. We therefore propose AdSeek, an agentic framework that dynamically selects visual and audio tools during multi-turn exploration, replacing static uniform sampling with active evidence acquisition. To address the sparse credit assignment of reinforcement learning, we develop an offline trajectory rectification mechanism that identifies reasoning errors and missing multimodal evidence in RL-generated trajectories. The corrected trajectories provide supervised fine-tuning signals that reduce biases learned during RL. This mechanism supports a rectified bootstrapping pipeline in which initial RL exposes reasoning bottlenecks, supervised fine-tuning corrects them, and a final RL stage further improves the policy. AdSeek achieves 74.30 percent accuracy on the AdsCVR test split, outperforming its Qwen3-VL-8B-Instruct backbone by 27.90 percentage points. It also generalizes to the open- domain CrossVid benchmark, demonstrating effective active evidence gathering.

Codon optimization, the process of selecting synonymous codons to improve mRNA translation efficiency and protein expression, is central to therapeutic protein production and mRNA vaccines, yet it remains a hard problem. The design space is discrete and combinatorially large, precluding gradient-based methods, and existing tools rely on heuristic proxies (e.g., Codon Adaptation Index or GC-content) that poorly capture true expression. We introduce Latent-Space Codon Optimization (LSCO), which recasts this discrete problem as a continuous one by mapping sequences into the latent space of a pretrained mRNA language model, enabling efficient gradient-based search. LSCO combines four components: a data-driven expression objective from an uncertainty-aware predictor, a Minimum-Free-Energy regularizer for structural stability, a naturalness prior from a protein-to-codon back-translation model, and constrained decoding for protein fidelity. On a real-world, wet-lab antibody expression dataset, LSCO outperforms simple frequency-based, as well as modern deep generative baselines in predicted expression, while retaining suitable biophysical properties.

When two AI agents disagree, who persuades whom? As multi-agent systems increasingly combine language models of different families and sizes, the answer can determine which judgments survive interaction. Measuring persuasion as the probabilistic shift in an agent's decision after a single exchange with a dissenting peer, we test seven open-weight models across three language understanding tasks. We find that persuasion is strong: when models disagree, receivers often abandon their initial judgment after seeing a peer's answer and explanation. Surprisingly, however, neither standalone certainty nor model scale reliably predicts persuasion dynamics. Models producing almost perfectly consistent decisions in isolation can be among the most susceptible to persuasion, and small models can match larger ones as persuaders and resist their influence just as effectively. Furthermore, we show that the size of the shift depends more on the susceptibility of the listener than on the persuasiveness of the speaker. Persuasion patterns are therefore specific to each model pairing, with heterogeneity amplifying persuasion in some combinations and suppressing it in others, allowing a dissenting agent running a small model to overturn the judgments of a much larger one. These findings show that the behavior of interacting models cannot be inferred from their individual properties but must be evaluated in the combinations in which they will operate.

Standard autoregressive (AR) models process high-level task instructions, state history, and transient tokens within a single shared sequence of tokens. Consequently, they lack the architectural mechanisms needed to isolate macro-objectives from context noise. To overcome this single-channel limitation, we introduce Latent-Steered Autoregressive (LS-AR), a dual-channel architecture that decouples continuous goal steering from discrete token decoding via FiLM conditioning. We evaluate a Static Goal Encoder (P_0) for persistent macro-objective retention across long rollouts and a Dynamic State Tracker (P_t) for recurrent latent updates during generation. On long-horizon retrieval past context limits (H=1024, W=500), LS-AR (Static) achieves 100% target recall where parameter-matched baselines collapse (0%), while increasing throughput by ~35% and cutting peak VRAM by 52.8%. In Blocksworld planning under forced perturbations (k=1), LS-AR (Dynamic) sustains an 89.0% completion rate vs. 71.0% for the baseline, though zero-shot entity scaling (N -> N+1) exposes single-vector capacity limits (0%). Finally, dual-channel authority analysis shows that text goal dropout establishes latent-dominant control, offering structural defence against text prompt injection while introducing a latent vector attack surface.

Scientific discovery often begins when scattered clues call for a new way of describing the world. Such abductive exploration can require constructing the representation in which an explanation is stated, when the world is epistemically open, and composing evidence scattered across contexts, when it is structurally interconnected. Existing benchmarks rarely separate these two demands or control them independently. We introduce ULTRADISCOVERY, an interactive world of five domains in which an agent revises an initially successful theory and predicts the outcome of an unseen cross-domain intervention. A $2 \times 2$ design leaves the representation open or discloses it, and leaves the evidence distributed or aligns it, with the latent dynamics fixed. With the representation open, agents across eleven models often retract the axiom they were taught, and none introduces the unobserved entity or rewrites the variables that a replacement requires. Disclosure triples intervention requests and adds about one of the eighteen findings the world affords, and alignment adds less. Two vendor-harness systems carry discovery into more domains, and one of them rewrites the variables in Open episodes. No system makes the exact prediction within 200 paid actions. At larger budgets one exact prediction appears with both aids, while every Open episode remains inexact. The results locate the difficulty in the step from accumulating evidence to composing it into a representation that transfers.

Modern Computer Use Agents (CUAs) directly interact with graphical user interfaces and execute third-party web tools, exposing them to indirect prompt injection across every rendered page and tool response. While the Dual-LLM pattern is the primary system-level architecture offering formal security guarantees - using an isolated Planner LLM (P-LLM) to fix execution paths before processing untrusted inputs via a Quarantined LLM (Q-LLM) - these guarantees break down in graphical environments. Because CUA interaction is inherently dynamic, plans cannot remain data-independent; they must branch based on anticipated runtime web content - covering all possible cases the agent may encounter. This exposes agents to branch steering attacks, where an adversary crafts untrusted data to coerce a CUA down a hazardous, pre-approved branch without injecting explicit instructions. We systematically study branch steering attacks and introduce STEER-Bench (101 tasks across 9 domains), showing high attack success against both standard (94.4%) and vanilla Dual-LLM (89.5%) CUAs. We then propose COBRA, an architecture that pairs trusted branching plans with ahead-of-time capability constraints, strictly bounding the parameters and destinations each branch may execute. On STEER-Bench, COBRA reduces attack success to 0% while retaining 97% benign utility.

Deterministic data loading is important for foundation model development: model researchers need confidence that differences they observe across costly ablations are caused by the parameter they changed rather than non-determinism in the training data sequence. The data loader must provide elastic determinism, i.e., a deterministic sequence of global training data batches despite changes to the GPU topology across runs (e.g., due to GPU scarcity), frequent checkpoint-resume cycles, and different data processing execution backends. Achieving this is difficult because modern foundation model data pipelines tokenize, pack, and mix samples online, introducing stateful n-to-m transformations that break sample indexing. Existing data loaders largely assume indexable 1-to-1 pipelines, and the common workaround of offline materialization is expensive and, for some modalities such as video, infeasible. We present Zephon, a data loader for foundation models that supports online, stateful pipelines while providing elastic determinism and efficient resumption from checkpoints. It partitions the global stream into topology-independent lanes, serializes ordering decisions while parallelizing stateless work on interchangeable backends, and checkpoints only bounded in-flight state so recovery cost does not grow with training progress. We evaluate Zephon on text and vision-language workloads and show that it achieves competitive throughput while providing a combination of guarantees that no existing loader offers for online, stateful pipelines.

Entropic Optimal Transport (EOT) has become a practical framework for learning stochastic couplings between complex distributions, with applications in generative modeling and domain adaptation. However, most EOT solvers are designed for Euclidean spaces, while manifold extensions remain limited and often rely on costly iterative methods, simulated dynamics, or generic neural models that do not fully exploit the underlying geometry. We introduce ManifoldLightOT, a light approach for learning kernel-induced EOT couplings directly on common manifolds. Using the kernel form of the EOT solution, we construct geometry-specific Gibbs kernels together with compatible potential parameterizations for spheres, tori, $\mathrm{SO}(3)$, and $\mathrm{SE}(3)$. These choices yield closed-form normalization and directly sampleable conditional distributions. Our formulation naturally extends to products of manifolds, making it applicable to more complex geometries. The parameters of the potentials are optimized directly from samples using Monte Carlo estimates of the learning objective. Through synthetic and real-world experiments, we show that ManifoldLightOT often outperforms existing manifold OT methods while retaining direct sampling.

Text-to-image (T2I) models are judged by benchmarks that measure whether requested content appears, but these benchmarks largely overlook the complementary ability to satisfy negated constraints, for example, generating "a non-red cup." Measuring negation raises challenges not faced by affirmation-based benchmarks and requires careful prompt and evaluation design. We introduce NegT2IBench, a benchmark of 4,800 prompts covering two attribute types and four relation categories. Prompts are organized by polarity: the number of positive statements that must hold and negated statements that must not, each ranging from 0 to 2. Varying the two independently separates the effect of negation from the effect of prompt complexity. Our detector-based scoring is reproducible, auditable, and pinpoints which requirement failed. On 600 images with three-annotator labels, it agrees with humans as closely as vision-language judges up to 30x larger, while using only a fraction of their GPU memory. Across eleven T2I models and 211,200 images, nine score lower on a single negated statement than on a single positive one. Per-statement scoring reveals that the loss is largest for color and near zero for proximity, and that 41.5% of failed statements render exactly what the prompt forbids. Rendering what a prompt asks for and withholding what it forbids are distinct capabilities that an aggregate compositional score cannot distinguish. NegT2IBench measures the latter directly, providing a controlled testbed for diagnosing negation failures and developing methods to overcome them.

Bridges contribute significantly to transportation connectivity and urban development. Therefore, reliable bridge monitoring is crucial for protecting public safety and detecting anomalous behavior in bridge sensor data that may provide early indications of abnormal structural conditions. This paper investigates anomaly detection in real-world bridge sensor data using two different complementary approaches, namely signal processing and the data-driven machine learning model Isolation Forest. The real-time bridge sensor data is collected from an iBridge sensor device installed on a bridge in Norway. The methods are evaluated using anomaly counts, anomaly detection time, processing rate, anomaly rates, visualization, and temporal agreement. Moreover, a controlled anomaly-injection analysis is performed to evaluate the sensitivity of each method. Numerical results demonstrate distinct detection characteristics and computational requirements, highlighting the potential of machine learning, particularly the data-driven Isolation Forest, alongside signal processing for identifying anomalies in bridge sensor measurements.

Large language models are moving from producing trading signals to writing the code that executes them. The failure mode of the second role is silent: generated code runs, a backtest plots, yet the risk logic that the trader described is not the logic being executed. Existing code benchmarks test functional correctness on unit tests and finance benchmarks test forecasting; neither measures whether an implementation behaves like the strategy that was asked for. We introduce MintEval, a benchmark in which reference strategies are generated programmatically from a library of composable building blocks, back-translated into colloquial trader instructions, and re-implemented by the model under test. Generated and reference programs are executed bar by bar on identical market data and frictions, and compared on their actions rather than on code similarity or profit: alpha is differenced away. MintEval v0 contains 800 tasks on BTCUSDT 15-minute data, stratified by an execution-measured state-span complexity tau that is decoupled from description length. Low-cost models reach a mean ActionMatch of at most 0.544 and reproduce at most 0.087 of tasks exactly; on a stratified subset of 200 tasks a frontier model (Claude Opus 5.5) reaches 0.889 and reproduces 0.575 exactly, yet still fails silently on 0.275 of tasks. Given a menu of building blocks, models identify the strategy almost perfectly, yet 79.2% of the implementations whose specification was read correctly diverge on more than 10% of active bars. The LLM judge of a recent strategy-generation benchmark, applied verbatim, accepts every one of these silent failures.

Genuine embodied agency requires robots to turn continuous real-world experience into lasting, transferable skills. This demands continual learning that integrates new capabilities without eroding prior knowledge as tasks and environments evolve. Experience replay mitigates forgetting, but storing complete demonstrations becomes costly as tasks accumulate. World-action models offer a generative alternative, reconstructing past experience through joint predictions of actions and future observations. However, visually coherent rollouts may contain actions that cannot realize the predicted transitions, while new-task adaptation can disrupt previously learned behavior. RIFAR therefore combines reliability screening with drift-aware replay selection. It reconstructs trajectories from compact demonstration prefixes and uses a frozen inverse-dynamics model to assess action-visual consistency. Training first combines current demonstrations with the highest-quality screened trajectories. RIFAR then compares action predictions before and after this adaptation on identical historical inputs, reselecting trajectories with larger normalized drift from the same screened pool for continued training. Across three LIBERO suites and real-world experiments, RIFAR surpasses the previous state of the art in WAM-based generative replay. On LIBERO-Goal, it achieves 90.97 AUC while retaining only 320 historical time steps per task, approximately 4.9% of the steps retained using 50-demonstration replay.

Quantifying conversational dynamics requires reliable identification of interactional units and their temporal boundaries, but speech activity alone does not distinguish conversational turns from listener feedback or within-turn pauses. We present an automated pipeline for extracting turns and backchannels from separate-channel recordings of spontaneous dyadic conversation, designed to provide a consistent first-pass annotation for subsequent human review. The pipeline combines voice activity detection, channel-energy filtering, temporal merging, automatic speech recognition, and context-based post-processing. We evaluated the pipeline on 99 ten-minute Danish conversations from 33 dyads using segment-level detection reliability and temporal boundary error. Conversations were recorded under both normal and asymmetric listening conditions. In the latter, speech-shaped noise was delivered to one participant through bone-conduction headphones. Overall detection reliability was F1=0.621, with similar performance for turns F1=0.624 and backchannels F1=0.618. For successfully matched segments, median absolute onset and offset errors were 0.150 and 0.160s for turns and 0.130 and 0.180s for backchannels, respectively. Mean errors were substantially larger for turn boundaries, indicating a smaller number of large boundary mismatches. Performance did not differ significantly across the two experimental listening conditions. In a four-conversation case study, pipeline-human agreement was lower and more variable than human inter-annotator agreement and varied across parameter settings. These results support the pipeline as an automated first pass within a semi-automated annotation workflow, providing a consistent basis for more standardised and reproducible annotation of conversational dynamics.

Propaganda detection is an essential task in natural language processing (NLP), particularly in the context of manipulative political communications. However, identifying specific propaganda techniques presents a significant challenge due to their often subtle nature and reliance on context, making them difficult to distinguish from legitimate persuasive language. Propaganda often involves highlighting certain facts while downplaying or ignoring others to create a desired perception. This biased communication aims to influence attitudes, beliefs, or behaviors towards a particular cause or position. This paper explores advances in detecting propaganda techniques through a comparative analysis of modern language models, using the SemEval-2020 Task 11 dataset. We evaluated both masked language models (based on XLM-RoBERTa or DeBERTa V3) and causal models (from OpenAI, Google, Mistral, Anthropic and Meta), employing two prompting strategies: base and chain-of-thought prompting. Our results demonstrate improvements over state-of-the-art models, with the best-performing MLM achieving an F1 score of 63.18 in technique classification and the best causal model achieving 63.62. We also observed that certain models excel in specific techniques, such as loaded language and name-calling, while struggling with others like bandwagon and black-and-white fallacy. These findings suggest that fine-tuning, ensemble modeling, and the use of larger datasets can further enhance propaganda detection capabilities.

Security workflows need models that turn complex observations and explicit policies into decisions. System One models introduced by Jev return typed predictions and probabilities; security specialization supplies the domain expertise behind those predictions. We introduce SecJev, to our knowledge the first family of Jev-like decision models specialized for security, spanning 0.8B to 9B parameters. Built on Kev's single-pass candidate scorer, SecJev learns Boolean, choice, and ordered decisions from text, telemetry, and observation histories. We develop SecJev-Corpus to unify source-label prediction and explicit-policy evaluation across 14 tasks and eight sources. It covers tool outputs, traffic, federated updates, consensus, authentication, and vehicle messages. Scene-weighted training adapts the models across these domains while preserving a shared typed decision interface. Security specialization improves every model in the family; SecJev-0.8B outperforms general Kev-9B by 20.51 percentage points in task-macro accuracy. Comparisons with answer-only generative fine-tuning show close accuracy and latency with lower peak inference memory. Tests on new source groups reproduce gains over Kev in prompt-injection and traffic decisions, with capture-dependent false alarms. We release adapters, decision heads, SecJev-Corpus, and training and inference code.

Chance-constrained programs (CCPs) optimize decisions under uncertainty by limiting the probability of constraint violation. Despite advances in traditional and learning-based approaches, optimizing non-convex or non-smooth objectives and adapting to different objectives under fixed chance constraints remain challenging. In this paper, we propose a \textbf{D}erivative-free \textbf{D}iffusion-based framework that \textbf{D}isentangles constraint modeling from objective optimization, termed \textbf{D$^3$Opt}. We learn the chance-feasible structure once, independently of any particular objective, by training a risk-conditioned diffusion model solely on constraint-filtered decisions and freezing it as a reusable prior for post-specified objectives. At inference time, we propose an annealed, particle-based Feynman--Kac correction along the frozen reverse diffusion process to optimize post-specified objectives using only function evaluations. This enables derivative-free optimization of non-convex and non-smooth objectives without objective-specific retraining. We prove that the correction preserves feasibility when this property holds for the frozen prior, and derive an optimization-error bound separating learned-prior coverage, finite-particle approximation, and finite-temperature effects. Experiments on linear Gaussian CCPs, objective-transfer tasks, and chance-constrained economic dispatch demonstrate effective optimization across smooth and non-smooth objectives, including non-convex cases, and objective generalization under fixed chance constraints without retraining.

GC--EI--MS is an important technique for analyzing volatile and semivolatile compounds in complex samples. However, conventional methods rely heavily on reference spectral library matching, limiting their ability to identify compounds absent from these libraries and to infer complete molecular structures directly from fragmentation information. Here, we present DiffGCMS, a spectrum-conditioned discrete graph diffusion model for de novo structure elucidation from GC--EI--MS, and further develop a framework that integrates DiffGCMS with second-stage reasoning by a large language model (LLM). In the first stage, DiffGCMS generates candidate molecular structures from input spectra; in the second stage, the LLM uses mass spectral information to validate, repair, and rerank the candidates and provides interpretable analysis of fragment-ion peaks. This framework can generate plausible molecular structures for compounds absent from reference spectral libraries and provide traceable evidence supporting its decisions. On a test set comprising 13,696 spectra from NIST 20, the generative model achieved Acc@1 and Acc@10 of 6.01\% and 15.76\%, respectively. On the test subset containing molecules with no more than 10 heavy atoms, LLM-assisted molecular graph repair and reranking increased Acc@1 from 21.28\% to 21.95\%, Acc@10 from 46.91\% to 47.99\%, and candidate validity from 91.04\% to 100\%. These results demonstrate that spectrum-aware postprocessing can correct errors produced by the generative model while providing auditable and traceable explanations for the final ranking.

Learning control from action-free recordings is challenging because intervention effects are unobserved and policies may exploit errors in reconstructed dynamics. We present a hierarchical model-based reinforcement learning framework that uses shared structure across related systems to learn system-specific control policies from action-free recordings. A hierarchical dynamical system reconstruction model captures shared dynamics and individual variation through low-dimensional embeddings. These embeddings are then reused to parameterize shared policy and value networks, linking differences in reconstructed dynamics to differences in control. Policies are trained entirely via simulation under an explicit intervention model with additive latent perturbations. Piecewise-linear recurrent neural networks enable mechanistic analyses of the controlled dynamics, while decoder-based constraints make the immediate effects of interventions interpretable in observation space and permit interventions on one modality while protecting another from direct manipulation. On Lorenz-63 and double-pendulum systems, hierarchical policies improve transfer over independently trained policies. On Lorenz-63, they also achieve a higher mean reward than repeated planning with the same reconstructed models, perform comparably to methods trained with controlled interactions, and generalize to systems absent from policy training after embedding inference alone. Applications to neural-behavioral recordings demonstrate suppression of predicted movement under constrained neural perturbations. Together, these findings show how shared dynamical representations support transferable control and mechanistic hypothesis generation from action-free recordings.

Large Language Models (LLMs) are prone to generating hallucinated content, which compromises their reliability in knowledge-intensive tasks. To address this challenge without sacrificing creativity, we propose HARPO, a reinforcement learning framework designed to jointly optimize faithfulness and creativity. HARPO incorporates a Hallucination-Aware Generative Reward Model (HA-GRM), trained via verifiable feedback, to assess both faithfulness and writing quality. A Selective Activation Mechanism (SAM) activates writing rewards only for outputs judged hallucination-free by HA-GRM, while a data curriculum progressively shifts training from creative writing to hallucination-centric tasks. On RAGTruth, our Qwen3-4B-based HA-GRM achieves a response-level F1 score of 78.08%, compared with 66.37% for the supervised fine-tuning baseline. Experiments on Qwen2.5 and Qwen3 models from 1.7B to 8B parameters show improvements in both faithful generation and writing quality. On Qwen3-4B, HARPO reduces the HA-GRM-judged hallucination rate on MultiHopRAG from 3.29% to 1.02%, while increasing the Arena-Hard-v2.0 creative-writing score from 16.95% to 27.54%.

Relational foundation models are increasingly pretrained on synthetic databases, yet downstream benchmarks reveal little about why one synthetic corpus produces a better model than another. In particular, strong performance may arise from realistic row-level statistics without the model ever learning to use relational structure. We study this as a data-attribution problem: which property of synthetic pretraining data induces relational computation? Using four Relational Transformer checkpoints trained with the same architecture, initialization, objective, and compute budget on corpora produced by four relational data generators, we trace a measurable property of the data to learned computation and downstream behavior. We hypothesize that relational mechanisms emerge when cross-table information is predictively necessary for the masked-cell pretraining objective. RelDiff exhibits by far the largest predictive gain from foreign-key-linked parents, and its corresponding model is uniquely sensitive to foreign-key interventions on unseen databases. This dependence survives a random-initialization control, grows monotonically with the fraction of corrupted links, and localizes to a serial cross-table pathway. Finally, disrupting the same mechanism during downstream inference removes RelDiff's advantage on relational tasks while leaving structure-insensitive models nearly unchanged. These results connect a property of synthetic training data to a learned mechanism and, through intervention, to downstream behavior.

Large language models are increasingly used as reasoning components in AI-driven materials Co-Scientists, yet the reliability of the resulting verification pipeline remains unclear. Metal-organic frameworks (MOFs) provide a particularly challenging setting because structures may appear under different identifiers, synthesis outcomes depend strongly on experimental conditions, evidence is distributed across heterogeneous sources, and some hypotheses require computation rather than literature alone. We introduce a diagnostic benchmark with four task families covering structural grounding, synthesis-condition verification, evidence-sufficiency verification, and MLIP-based computational verification. T-MOF-1-3 are evaluated under closed-book, retrieval-enabled, and oracle-evidence settings to localize failures in knowledge access, evidence acquisition, and reasoning, while T-MOF-4 separately evaluates computational verification. Guided by these diagnosed failure modes, we develop MOF-Verify, a failure-aware agentic harness that targets structural, literature, evidence-sufficiency, and computational bottlenecks before producing a final verdict. Across multiple backbone LLMs, MOF-Verify substantially improves hypothesis-verification performance over direct inference and retrieval-based baselines. Benchmark datasets are released at https://github.com/IMMS-Ewha/MOF-Verify-Benchmark.

A coding agent can earn a passing grade by fixing its code, or by deleting the test that exposes the bug. Detecting such reward hacking requires recognizing attempted shortcuts, including those that fail. We release 173,561 annotated multi-turn coding trajectories from Qwen3-8B and show that supervising shortcut behavior independently of exploit success substantially improves detection. We introduce HACKTRACE, a behavior-supervised monitor that reads the internal states the agent already computes while generating code. Reusing these states enables monitoring before a turn is complete, without additional language-model tokens or passes. Combining this evidence with static features of the final files achieves a mean per-problem AUC of 0.997 with 8 ms of monitoring overhead, improving both accuracy and latency over monitors that run the model again on an honesty question and answer. The same generation states also provide an inexpensive monitoring signal for reinforcement learning. With strong GRPO penalties, HACKTRACE reduces the cheating share of passing solutions from 82-91% to 1-5%, while retaining honest, correct solutions and maintaining high detection accuracy as the policy evolves. Our results show that both the supervision target and the source of monitoring evidence matter for turning accurate detection into a useful training signal.

Clustered Federated Learning (FL) partitions a client population into groups of similar local distributions and trains one specialized model per cluster, mitigating client drift that degrades single-model methods under non-IID data. Prior methods discover cluster structure inside the training loop through gradient similarity, loss evaluation, or EM-style updates, thus increasing communication overhead, exposing gradients to inversion attacks, and providing no mechanism to assign clients absent from training. We propose RIPPLE, a clustered FL framework in which cluster assignment is computed entirely offline from a spectral characterization of each client's local data: a variance-weighted principal-component prototype embedded via the Wavelet Scattering Transform and decoded by a Gaussian Mixture VAE trained server-side on synthetic client populations before federation begins. Per-round communication cost matches FedAvg exactly, and a client absent from training obtains a personalized model from a single forward pass, without gradient computation, model evaluation, or extra communication round. We prove that the gap between RIPPLE's surrogate clustered objective and the oracle is bounded by a computable quantity decaying with client sample size and independent of federation duration; per-cluster convergence matches the minimax-optimal rate for non-convex smooth objectives. Across five benchmarks spanning controlled and realistic heterogeneity, RIPPLE consistently outperforms all baselines, with margins growing on the most realistic partitions.

Large language models (LLMs) contain broad knowledge, but they cannot access all of it reliably. We study this problem through knowledge accessibility, which describes whether the knowledge needed for a query can be recalled from the model. We find that knowledge accessibility has a simple geometric structure in the model's representation of the query alone, before any generation. More accessible queries are closer to a center in the representation space, while less accessible queries are farther away. This geometry reveals a knowledge boundary that separates more accessible queries from less accessible ones. Accessibility consistently decreases with distance from the center, and this distance-based ordering transfers across datasets even when the centers differ. Controlled experiments further show that the centered geometry is more closely related to knowledge accessibility than to reasoning difficulty. The geometry also reveals when different interventions are useful. Query rewriting helps more for accessible queries, chain-of-thought reasoning helps more near the boundary, and retrieval gives larger gains beyond the boundary. These findings not only provide a new geometric view of how knowledge is organized in language models, but also suggest a useful pre-generation signal for adaptive inference.

Long-form thinking traces can substantially improve the multi-step reasoning performance of large language models (LLMs), but they introduce high inference-time overhead, with latency dominated by sequential decoding. We propose HyperThink, a text-to-parameter approach that amortizes this reasoning computation into a single query-conditioned parameter update: a lightweight hypernetwork reads the question and predicts updates to a small subset of the base LLM's parameters, while a vector-quantized decoder constrains them to a finite set of reusable patterns to improve robustness and transfer. Trained end-to-end on outputs from the base model itself, HyperThink eliminates long thinking traces at test time: after one hypernetwork forward pass, the adapted model generates a concise step-by-step solution and final answer without an intermediate trace, using far fewer tokens while retaining strong reasoning performance. Empirically, HyperThink improves the low-latency region of the accuracy-latency trade-off on mathematical and general reasoning tasks, with its strongest gains in the near-non-thinking regime.

We present WebFovea, a vision-based web agent that placed 2nd in the WebRetriever Challenge 2026 with a final score of 57.0 out of 100. The challenge evaluates agents end to end on Protocol III of the WebRetriever benchmark (arXiv:2607.06118): starting from an entry URL on a live website, the agent must operate the site's own interface and return a verifiable answer. A capable multimodal large language model (LLM) is necessary for this, but not sufficient. The model's decisions reach the browser through the harness, the code between the model and the page. At every step, four things must go right: the model's reply must be parsed into the intended action, the action must take effect on the page, the result must be reported back accurately, and the model must be shown the information it needs. On real websites, many of the failures we observed occurred at one of these four stages rather than in the model's reasoning. A coordinate-space mismatch placed every click at 3/4 of its intended coordinates; actions on native dropdowns, inside iframes, and in text boxes failed silently; and self-generated chat-template tokens contaminated 4.9% of task episodes. WebFovea hardens each stage and surrounds the loop with guardrails that keep the agent within the rules and its budget. The four-stage view does not depend on the model, although some individual fixes do. Because we used the same model in all four submissions, the rise of our official hidden-set score from 31.0 to 57.0 reflects changes to the harness, up to run-to-run variance on live sites. We describe the design, the evidence for each component (including negative results), a failure analysis, the limitations, and a roadmap that includes routing different steps to different models.

Multimodal learning often suffers from modality imbalance, where the joint optimization process is dominated by a single modality. Existing methods typically estimate modality imbalance from score disparities derived from prediction uncertainty or optimization statistics. However, due to distinct prediction uncertainty and learning dynamics across modalities, direct comparison of such scores may misinterpret intrinsic modality differences as progress gaps, leading to biased imbalance estimation. In this paper, we propose Function-Space Guided Multimodal Optimization (FGMO), which leverages a function-space progress signal to assess modality-wise optimization progress and coordinate optimization across modalities to alleviate modality imbalance. Specifically, we introduce Functional Progress Estimation (FPE) to measure each modality's update-induced function-space response and calibrate it against a loss-aligned unimodal reference, producing a comparable progress signal. Based on this signal, Functional Response Control (FRC) redistributes modality-level function-space budgets and realizes the target responses through tensor-wise learning-rate adjustment. Theoretical analysis establishes a one-step target-contraction property of FRC under bounded controller-state mismatch, and extensive experiments demonstrate the effectiveness of FGMO across multiple multimodal benchmarks.

Diffusion models rely on numerical solvers requiring time-discretization, which has a large influence on the tradeoff between sampling cost and quality. However, the computational difficulty of the reverse process varies along the sampling trajectory and across data distributions, making the choice of discretization important. We adapt proportional-integral (PI) step-size control to diffusion, using our diffusion noise-normalised error estimator. Unlike existing adaptive methods in diffusion that respond only to the current error, the PI solver also incorporates the previous error, yielding smoother step adaptation. We further show that these per-sample trajectories exhibit shared structure and can be aggregated into a fixed schedule that retains much of the benefit of adaptive sampling. We evaluate both approaches on natural-image and language datasets, in terms of quality, measured by FID at a matched number of neural network evaluations (NFE), comparing them with widely used stochastic solvers and schedules. For images, our fixed discretization outperforms the commonly used EDM schedule in terms of sample quality when used with the stochastic Heun sampler, and with the EDM-churn sampler at low NFE. Additionally, our PI adaptive solver obtains better FID than most stochastic and adaptive baselines, although it does not beat the EDM-churn sampler at low NFE. Moreover, we find our solver outperforms both the EDM and the entropy schedule on language diffusion at low-to-medium NFE in terms of perplexity, with the drawback of lower token entropy. Lastly, we find that the benefit of per-sample adaptivity is problem-dependent. It is highly beneficial in 1D toy examples, while only marginal for image and language data, where the average schedule sometimes even outperforms the PI-adaptive solver. Code is available at https://github.com/ellakemperman/adaptive-second-order-diffusion-solvers

Large language model (LLM)-based agents increasingly operate in multi-agent systems (MAS) characterised by strategic interaction. However, little is known about whether, and to what extent, different types of messages affect the outcomes of strategic games. By investigating AI agents based on four popular LLMs, playing four games with different cooperation equilibria, we study whether messages of different kinds (natural language, numerical signals, or random sequences) significantly modify the levels of cooperation in each game, also depending on the agents' assigned personalities. We observe that structured messages alter the final payoffs for most games and LLMs, but without a predictable pattern; this challenges the assumption that AI agents can converge to stable equilibria regardless of additional capabilities. Moreover, we observe that agent-generated numerical messages depart from randomness, most strongly and consistently when agents are explicitly instructed to communicate; however, they introduce an additional interpretability challenge, as their symbol distributions are mostly associated with the payoff structure and typically become more concentrated with repetition, but are overall difficult for humans to interpret. Monitoring for coordination of AI agents through restricted channels should thus prioritise message-level fingerprints, which generalise across models, over behavioural decisions, which do not.

Gene set analysis is a cornerstone of functional genomics, yet it remains labor-intensive and heavily dependent on manual curation and expert biological interpretation. While Large Language Models (LLMs) have emerged as powerful tools for genomic reasoning and annotation, most existing approaches rely on symbolic gene names and fail to capture domain-specific biological structure, particularly protein sequence information that governs molecular activity, interactions, and downstream gene function. In this work, we propose SoftGene, a novel framework for LLM-based gene set annotation that leverages the hierarchical structure of gene sets. First, we use a hierarchical attention-based encoder built on ESM, a protein language model, to represent each gene set using protein-level amino acid sequence information. Second, we construct a hybrid prompting scheme that combines soft prompts derived from gene set embeddings with hard prompts containing auxiliary context generated by an LLM, and feed the resulting prompt into a local LLM for annotation. We evaluate our framework on two benchmark datasets: Gene Ontology (GO) and the Molecular Signatures Database (MSigDB). Our results show that integrating protein-sequence representations with textual context improves gene set annotation overall, while per-domain analyses reveal that the contribution of protein embeddings varies across biological domains.

The key-value (KV) cache becomes a major memory bottleneck in long-context LLM inference, placing substantial pressure on memory capacity and bandwidth. To mitigate this bottleneck, vector quantization (VQ) has emerged as a promising approach for aggressive KV cache compression. However, existing VQ methods degrade substantially in the 1-bit regime. At such extreme compression, each codebook must represent a larger group of channels with a limited set of centroids, making effective use of its capacity increasingly challenging. To address this, we introduce $\textbf{TaSQ}$, which tailors the VQ target space by combining query-guided channel weighting, cross-head normalization, and covariance-aware channel grouping to better reflect the error sensitivity and statistical structure of cached activations. Since these transforms are RoPE-compatible and can be easily merged into projection weights and codebooks, TaSQ preserves the conventional VQ lookup structure and adds negligible serving overhead. Across general, long-chain-of-thought reasoning, and long-context retrieval benchmarks, TaSQ consistently outperforms existing low-bit KV cache VQ baselines while preserving reasoning stability. On a single RTX 6000 Ada GPU, its SGLang implementation supports up to $14\times$ larger batch sizes and achieves $1.87\times$ higher peak throughput compared to the BF16 baseline.

What makes a short story gripping; a news article newsworthy; or a math proof elegant? These constructs resist articulation or verification; their meaning is at least partially tacit. However, modern AI models are improved primarily via articulated constitutions, rubrics and verifiers (i.e. in RLAIF and RLVR); tacit components of preferences are typically understudied. We introduce a large, labeled preference dataset CreativePreferences, containing 2.8M texts labeled by 317M human preference judgments across 7 creative domains, with 42 benchmark tasks. We model these labels with executable programs, rubric banks and densely trained models (V, A and VAT, respectively). We observe robust articulability gaps, VAT-VA; and verifiability gaps, VAT-V; we estimate upper and lower bounds for each gap with a novel measurement approach that discovers articulable and verifiable metrics, identifies spurious variables and estimates the value of undiscovered metrics using capture-recapture. These gaps occur across all domains, even in domains traditionally treated as fully verifiable: correctness-centered domains (i.e. mathematics and software engineering) and claim- and novelty-centric domains (i.e. news, patents, peer review). The size of the gap varies based on domain (e.g. peer review and creative writing have the largest articulability gaps) and widens as more people take part in the judgment, consistent with Collins' collective tacit knowledge. We show two consequences: (1) on human generations, the full model more closely matches human preferences, often in disagreement with articulated criteria, and (2) in an analogy to Goodhart's law, articulating preference shifts it away from the tacit dimension. Articulability and verifiability gaps are consequential; we give recommendations on when tasks can be prompted; how learning mechanisms might improve; and when to leave judgments with humans.