The Inference Report

Native visual reasoning treats visual generation as the medium of reasoning itself: visual states (i.e. images and videos) are not merely inputs to be understood or outputs to be rendered, but first-class substrates for problem solving beyond language. Yet progress remains bottlenecked by the lack of scalable training tasks, reliable feedback, and controlled comparisons across generative substrates. In this work, we introduce VBVR-Pro, a closed-loop testbed that makes native visual reasoning through generation trainable, verifiable, optimizable, and experimentally controllable. 1) Task scaling. VBVR-Pro turns visual reasoning into a controlled task space of 300 procedurally generated tasks. Models trained on VBVR-Pro show strong transfer beyond the proposed suite across seven external visual reasoning benchmarks such as RISE-Video, MME-CoF-Pro, and BabyVision. 2) Verifiable rewards. VBVR-Pro provides verifiable reward scorers for task-grounded evaluation. Through a systematic study of leading MLLMs as judges, we identify recurring failure modes of the prevalent VLM-as-a-judge paradigm. In contrast, the proposed scorers are grounded in deterministic, task-specific rules, achieve fine-grained alignment with human judgments. Importantly, they serve as reliable reward signals for large-scale multi-task reinforcement learning and demonstrate stronger post-RL performance across visual reasoning tasks. 3) Mechanism study. VBVR-Pro enables controlled modality studies across more than 30 image, video, and interleaved generators. Our analysis shows that video generation remains strongest for tasks requiring persistent spatiotemporal state tracking, while interleaved generation provides a compute-efficient alternative. Critically, ablations and probing suggest the presence of vision-native trajectories that are crucial to visual reasoning. We release all data, models, scorers, and code.

In this paper, we explore a novel task of Multimodal Unsupervised Continual Post-Training (MU-CPT), enabling deployed MLLMs to continually evolve from streaming unlabeled data. Existing unsupervised post-training methods for MLLMs typically optimize target tokens uniformly, overlooking their heterogeneous visual dependence (VD). However, we reveal that token-level VD is crucial for MU-CPT. Specifically, its structural distortion serves as an indicator of cross-modal catastrophic forgetting, and its inherent heterogeneity acts as a compass to guide new-task learning. Leveraging this property, we propose a Visual Dependence-Aware (VDA) framework with two main components. First, Visually Constrained Optimal Transport (VC-OT) formulates the VD structural distortion of old-task VD during new-task learning as an optimal transport problem to mitigate cross-modal forgetting. By designing a region-aware ground cost and a dependence-stratified transport penalty, it prevents global shifts in visual focus while strictly prohibiting visual reliance from degenerating into language bias. Second, Visually Modulated Adaptation (VMA) exploits VD heterogeneity to emphasize visually grounded new-task learning, promoting new-task plasticity. Together, our method simultaneously maintains old-task stability and new-task plasticity during challenging MU-CPT. Extensive experiments under our MU-CPT setting validate the effectiveness of VDA.

Existing action quality assessment (AQA) datasets and methods rely primarily on visual inputs such as RGB and pose, overlooking physiological dynamics such as muscle mechanics and often modeling actions as monolithic patterns. These limitations hinder fine-grained, biomechanically grounded feedback. We introduce MyoMechanix, a multimodal ecosystem for weight-loaded actions that aligns motion with muscle activity. Expert-annotated, it contains 7,500+ samples of 20 actions from 38 subjects, with synchronized multiview RGB video, 3D pose, sEMG, and additional physiological signals, forming the largest multimodal AQA benchmark to date. We further construct the Fitness Knowledge Graph (FKG), which organizes expert annotations into structured relationships among actions, phases, key steps, errors, and corrective feedback, enabling compositional scoring and interpretable assessment. Building on these representations, we develop CUBIST (Compositional Ontological Reasoning Engine), which performs decomposition-analysis-recomposition for fine-grained error attribution and feedback generation. We also establish MyoMechanix-AQA, MyoMechanix-VideoQA, and a novel MyoMechanix-Video2EMG task. Experiments show that multimodal sensing and structured representations improve performance, interpretability, and error attribution, with CUBIST achieving state-of-the-art results; VideoQA enhances language-grounded action understanding; and Video2EMG suggests video-based alternatives to costly EMG sensing. MyoMechanix advances skilled activity understanding toward biomechanically grounded, multimodal, and compositional reasoning for Physical AI applications in fitness, rehabilitation, healthcare, and machine learning. Project page: https://haoyin116.github.io/MyoMechanix/

Designing machine learning algorithms for wireless resource management is labour-intensive: the architecture, the loss function and the training recipe are all specified by hand. We demonstrate that this design layer can be surrendered to an autonomous agent in its entirety. We adopt the autoresearch protocol, in which an AI coding agent edits a training script, runs a fixed-budget experiment, and retains or discards the change according to a single immutable metric. We grant the agent authority over the architecture family, the input representation, the output parameterization, the loss function and the task-sampling law, and set it a target chosen for its difficulty: sum-least-percentile-rate power control across a multicell network. The formulation targets cell-edge throughput and is non-convex, non-smooth and strongly NP-hard away from its max-min vertex. Safeguards render the results trustworthy: a hash-pinned evaluator, an enforced inference contract and a pre-registered falsifier per experiment. In eighty-one unattended experiments over twenty-six hours, the agent reached $99.5\%$ of a converged minorization-maximization reference in one fixed-cost inference pass, at roughly $600\times$ lower inference cost, closing $94\%$ of the gap from its first working architecture, with one parameter set serving every network size and percentile target. It recovered provable structure rather than tuned constants: the output parameterization it discovered reproduces the exact max-min-optimal allocation at the minimum percentile, for every value of the trained weights.

Civil infrastructure compliance checking has long relied on engineers manually reading legacy 2D plans; however, OCR-based automation strips away the geometry and layout essential for interpreting these plans. We present a Visual-First Multimodal Retrieval-Augmented Generation (RAG) framework called PlanSightRAG. It indexes and reasons directly over plan imagery, integrates a ColNomic-3B multi-vector retrieval, an agentic Planner-Retriever-Auditor-Synthesizer, and MaxSim heatmaps as an evidence trail. We introduce a 4,056-pair benchmark from five state Departments of Transportation (DOT) standard plans (1,898 pages). PlanSightRAG achieves 91.47% Recall@5 on zero-shot retrieval, while on a held-out Michigan DOT corpus, it achieves 91.40%. On synthetic, parametrically-generated compliance drawings, our Qwen2.5-VL-72B pipeline reaches 100% verdict accuracy only when supplied a pre-resolved rule threshold, a controlled ceiling that a non-VLM OCR baseline already reaches at 76.4%. Finally, we demonstrate autonomous visual rule-grounding by extracting numeric limits directly from a specification corpus without any human-supplied rules.

We present a first application of sparse-autoencoder-based mechanistic interpretability to particle physics. Studying a neutrino foundation model pretrained on IceCube data and fine-tuned for direction reconstruction, we identify a validated atlas of physical concepts in the model representation, using a strict validation protocol consisting of held-out tests, matched nuisance controls, and replication across independent dictionary trainings. Causal interventions show that the direction head barely draws on this atlas. Motivated by this underused information, we train an uncertainty head on the same event-level representation to predict the model's angular reconstruction error. Unlike the direction head, it depends causally on quality and brightness features from the atlas. At $20\%$ selection efficiency, this interpretable estimator improves the median angular resolution from $20.2^\circ$ to $3.2^\circ$. These results suggest that mechanistic interpretability can reveal learned latent physics encoded within a model's internal representation and help design downstream tasks that exploit it.

Addressing critical global challenges, from food security and disaster risk to disease outbreaks and socio-economic vulnerability, demands high-fidelity geospatial modeling. However, building predictive planetary models remains bottlenecked by a fragmented data ecosystem, requiring manual data retrieval, multimodal data curation and fusion along with iterative model selection. We present the Planetary Prediction Engine (PPE), an autonomous AI system that executes this end-to-end workflow directly from natural-language queries. PPE synthesizes multimodal datasets on the fly, retrieving spatiotemporally relevant covariates across open-web and Earth observation platforms (Data Commons, Google Earth Engine) and fusing them with geospatial foundation model embeddings (PDFM, AlphaEarth). Simultaneously, it searches over task-tailored model architecture families with automated overfitting guards. Across diverse tasks, geographies, and scientific domains, PPE consistently outperforms state-of-the-art or manually tuned expert baselines. For US spatial regression, PPE improves mean $R^2$ across 21 CDC health indicators (76.8% vs. 60.0%), FEMA national risk indices (64.9% vs. 60.0%), and the Social Vulnerability Index (66.2% vs. 58.6%). For spatial downscaling in data-scarce settings, PPE integrates localized proxies to double baseline accuracy in Nigerian food security indicators ($R^2$ of 66.1% vs. 31.5%). For epidemiological nowcasting of the 2026 DRC Bundibugyo Ebola outbreak, PPE achieves a Recall@10 of 83.3% (identifying 15 of 18 newly invaded health zones across five weekly forecasts), a +10.3 percentage-point improvement over the public state-of-the-art modeling (~73%). By combining autonomous multimodal planetary data discovery with targeted model optimization, PPE lowers the technical barrier to planetary-scale analytics, enabling rapid, customized, expert-level deployment.

Large language models write correct code for isolated problems but remain far weaker at autonomous machine-learning development, where an agent must revise data pipelines, models, and validation over hours of feedback, and on most competitions still finishes below strong human competitors. Outcome-based benchmarks record this gap but not its cause, because they grade the final submission and discard the development process behind it. We introduce TraceML, which pairs human and agent work on the same competitions under one version-level schema: 4,465 human Kaggle trajectories across 134 competitions, seven of which are also worked by two agent scaffolds, giving 430 paired human and 207 agent trajectories. Every code version carries its score, its timestamp, and labels for the action taken, its intent, the edit size, and the score effect. Read this way, the gap becomes concrete. Experts alternate data work, validation, model changes, and ensembling, and return to approaches they had set aside. Each agent scaffold instead collapses into a narrow loop: Codex spends its steps re-weighting ensembles and tuning submissions, MLEvolve mutates its model in place, and neither pivots at the human rate nor reopens abandoned work. A short planning prompt distilled from human practice moves the behaviors it names toward the human profile and lifts scores, but the effort profile stays agent-shaped: instruction closes only the part of the gap that reduces to instructions. We release the corpus, the schema, the labelers, and the extraction pipeline at https://huggingface.co/datasets/jerryyan/TraceML.

Deep neural networks often exploit spurious associations in their training data, a failure known as shortcut learning. Concept-based explainability methods screen for shortcuts by testing whether concepts such as a patient's sex or scanner settings can be decoded from a network layer. Because each concept is evaluated in isolation, these methods can mistake correlations between concepts as evidence that the model uses them. We introduce ICON decomposition, which instead quantifies how much of a layer's variance each concept explains after accounting for all other concepts and the outcome. On synthetic data with known ground truth, ICON recovers concept importance more accurately than seven alternative baseline methods. On skin-lesion and brain-imaging models, it isolates the concepts on which a model genuinely relies, quantifies the representation unexplained by any of the supplied concepts, and yields sparse explanations that we validate by retraining and out-of-distribution testing.

Collective intelligence can emerge when individuals coordinate through a shared environment, allowing local actions to accumulate into durable social organization. Language-model agents offer a new substrate for this process, yet most multi-agent systems rely on direct conversation, predefined roles, or centralized workflows. It remains unclear whether decentralized agents can build functional technologies and outperform independent search. Here, initially homogeneous LLM agents in SwarmWorld self-organize without assigned roles or recipes into evolving technological societies. Agents explore a spatial environment, process resources, test materials, construct persistent artifacts, and write executable controllers evaluated by a deterministic simulator under unseen disturbances after the agents are removed. SwarmWorld splits cognition from consequence: agents propose architectures and controllers within fixed action and material schemas, while the simulated world determines function. Shared societies develop broader, more resilient technological portfolios than a strong best-of-N isolated-search baseline, although isolated search remains competitive for the strongest artifact. Agents differentiate into exploration, construction, maintenance, and coordination behaviors, transitioning as the world matures. Technologies accumulate through collaborative construction, executable inheritance, and persistent agent-artifact networks, with most reuse beginning through physical observation rather than communication. Explicit cultural mechanisms amplify collaboration and organization, but functional benefits depend on outcome and timescale. Physical stigmergy alone supports capable societies, while interaction drives persistent technological ecologies rather than universally superior individual inventions.

Intent misinterpretation during vehicle interactions causes recurring planning failures. We study a decision layer in which a language-guided intent module reads structured descriptors, computes a smoothed intent-geometry divergence score, and gates the planned maneuver before commitment, upstream of a corridor envelope. On a replayed off-road departure and four crash clips under a frozen, disclosed implementation, gating is the only layer that repairs the plan: on the main case it fires 72 ms after the drift onset but 161 ms before the corridor exit, keeping the trajectory in the corridor in all ten replays. The first calibration draws nine false triggers in 5.9 minutes, each from scoring uncertainty as half a conflict; a preregistered redesign treating uncertainty as abstention cuts this to 0.341 per minute. Two ablations bound the model's contribution: the full score detects fastest on four of five failures under the deployed eligibility, three of five against the unvetoed rule (000871 by one cycle; 000228 by a pre-onset fire on an uncertain stretch that five clips cannot classify as signal or coincidence; dropping the confidence term costs two detections), while on in-domain tracks at equal false positives the geometric rule more than triples its detection. The evidence supports the gating mechanism; the model's demonstrated roles are the fastest detection on these failures and an uncertainty veto on the geometric rule.

Test-time scaling uses extra test-time compute to improve performance, such as letting language models reason longer when solving a problem. As models keep the entire reasoning trace in memory via full attention, hard tasks that need long thinking can be prohibitively expensive. However, we find most intermediate reasoning tokens lose importance as the model continues reasoning. This calls into question whether retaining them is worth the cost. Based on this insight, we propose Prefix Sliding, which discards tokens during reasoning that are not part of the prefix or the window of the last few thousand tokens. The prefix has key instructions and tools available to the model, while the most recent tokens are the current reasoning the model is working on. This caps the total memory requirement regardless of how long the model reasons, allowing for efficient long-horizon test-time scaling. Without training, Prefix Sliding can make existing models 3x faster while maintaining performance. Training with Prefix Sliding using reinforcement learning can achieve better performance by enabling scaling to reasoning traces beyond a hundred thousand tokens. Ablations show Prefix Sliding outperforms summarizing intermediate tokens or vanilla sliding window. Our code is at https://github.com/Muennighoff/prefix-sliding

Large-kernel Convolutional Neural Networks (CNNs) deliver remarkable performance in vision tasks by significantly expanding receptive fields, yet their quadratic parameter growth critically impedes storage-efficient edge deployment. While existing efficient architectures adopt parameter-efficient depthwise separable convolution backbones that leverage techniques like low-rank approximation and weight sharing to compress depthwise convolutions, we identify a critical oversight: pointwise convolutions dominate parameter volume (>87% in models like RepLKNet-31B) and constitute the primary deployment bottleneck on resource-constrained edge devices. This results in prohibitive storage costs and severe memory-loading constraints on resource-limited devices (e.g., smartphones with 4-12 GB Random Access Memory (RAM)). To overcome this, we propose Channel Group-Shared (CGS) low-rank approximation, a novel Singular Value Decomposition (SVD)-based parameter-sharing strategy. CGS constructs a structured low-rank paradigm isomorphic to SVD decomposition, comprising shared (high-parameter-cost) down/up-projection matrices across channel groups within a layer and channel-group-specific (low-parameter-cost) scalable diagonal matrices. This group-sharing design achieves significant parameter reduction. Extensive experiments demonstrate that large-kernel CNNs (RepLKNet, ConvNeXt, SLaK) enhanced with CGS strike an empirically favorable balance between competitive performance and substantially reduced storage costs. Crucially, by alleviating storage constraints, reducing memory bandwidth pressure during loading, and minimizing model loading latency, CGS enables the feasible deployment of pre-trained large-kernel CNN models on edge devices, thereby bridging the gap between high-performance vision models and practical edge deployment.

Automatic Speech Recognition (ASR) technologies have achieved remarkable performance in recent years through the use of large multilingual foundation models. However, most advances remain concentrated on high-resource languages, while indigenous languages continue to suffer from a lack of speech resources and language technologies. This work presents a preliminary study on the adaptation of Whisper for Automatic Speech Recognition in Baniwa, an indigenous Arawakan language spoken in Brazil, Colombia, and Venezuela. The experiments were conducted using a corpus of 1,373 manually transcribed recordings obtained from a linguistic documentation project. The corpus contains approximately 0.54 hours of speech and consists primarily of isolated words and short elicited utterances. The Whisper Small model was fine-tuned using supervised learning and evaluated using Word Error Rate (WER) and Character Error Rate (CER). The best model achieved a WER of 37.5% and a CER of 7.45%, demonstrating that multilingual foundation models can be successfully adapted to extremely low-resource indigenous languages. The results establish an initial baseline for Baniwa Automatic Speech Recognition and provide a foundation for future research involving larger datasets, language-specific adaptation strategies, and post-processing techniques.

Reasoning in language allows foundation models to spend more test-time compute on hard problems, such as those requiring decomposition, constraint tracking, and prediction of future consequences. Whether this mechanism can improve robotic manipulation remains unclear, where long-horizon tasks require tracking partial progress, reasoning about object relations, recovering from mistakes, and steering noisy low-level policies. In this paper, we study whether VLMs can be trained to reason directly in natural language to guide low-level manipulation policies. We introduce $R^3$, a simple post-training recipe that turns off-the-shelf VLMs into robotic reasoners: it first mid-trains a VLM on expert-generated reasoning traces to initialize the desired reasoning style, then improves the reasoner with single-step rubric-based RL from offline action data. Unlike prior robotic reasoning methods that mostly use structured traces as auxiliary supervision, $R^3$ trains free-form language reasoning to produce test-time guidance for action. We instantiate $R^3$ on Language Table and simulated bimanual grocery packing, two controlled testbeds for studying robotic reasoning and long-horizon manipulation. $R^3$ improves exploration and generalization across unseen tasks and significantly outperforms instruction-only imitation learning baselines on both benchmarks. Our analyses suggest that free-form language reasoning can function as a test-time compute mechanism for steering low-level policies. Our project page is available at https://robotic-reasoner.github.io/.

Choosing the rank of a low-rank adaptation (LoRA) update is usually an empirical task. In this paper, we provide a task-dependent theory of the approximation error achievable at each LoRA rank for Transformer attention. We fix a pretrained attention head, a target attention function, and a distribution over inputs from the downstream task, and bound the smallest expected Kullback--Leibler (KL) error achievable by a rank-$r$ query LoRA update. When target attention probabilities are bounded away from zero, we prove a lower bound of the error proportional to $ψ(\|d\|_2)$, where $d$ is the difference between candidate and target attention scores and $ψ(t)=\min\{t^2,t\}$. We also prove an unconditional upper bound $\min\{\|d\|_2^2/4,\sqrt2\|d\|_2\}$. Under explicit realizability, geometry, and moment conditions, we then bound the best rank-$r$ error between an explicit multiple of $ψ(\sqrt{T_r})$ and $\min\{T_r/4,\sqrt{2T_r}\}$, where $T_r$ is the downstream-weighted tail energy of the target update. We also provide target-Fisher bounds when candidate scores remain within a fixed range of the target scores, and an unrestricted lower bound when a subset of tokens carries most of the probability mass. These spectral bounds describe finite-score approximation. We then construct explicit families in which softmax saturation makes the rank required to match the attention function strictly smaller than the rank required to match the finite logits. Finally, we extend the analysis to fused multi-head LoRA and joint query/key updates, exposing the effects of rank sharing and query/key factorization constraints.

We consider optimization applications with unknown parameters where the decision maker believes that the optimal value of the nominal problem-the optimization problem they would have solved if the true parameters were known-is unlikely to be large. This belief derives from information that humans have that is not captured in datasets, obtained from domain knowledge and interacting with the physical world. We propose an approach to evaluating policies that provides tighter performance guarantees if the decision maker's belief happens to be correct. Our main result shows that if computing a policy's worst-case performance is a convex program, then the value of human expertise-the maximum improvement in performance guarantees that can be obtained from the belief about the nominal problem-is equal to the minimax gap of a max-min problem. We illustrate our developments in assortment optimization and shortest path problems.

Multi-norm adversarial defense aims to protect neural networks against perturbations defined by different norm constraints, but existing methods typically optimize competing robustness objectives within a single parameter configuration, leading to substantial training cost and unfavorable robustness trade-offs. We propose Robust CurveMoE, an efficient mixture-of-experts framework that connects models specialized for different perturbation norms through a low-loss path and exploits the complementary robustness profiles of models along this path. Robust CurveMoE derives clean and norm-specialized experts from robustness-constrained curve locations and selectively expertizes only influential layers, while sharing the remaining parameters across routing paths. To further reduce curve-construction cost, we introduce contribution-guided partial updating, which selects influential curve parameters using initialization-based gradient scores. We also theoretically bound the objective gap between partial and full curve optimization. Experiments on CIFAR-100 and ImageNet-100 with WideResNet and Vision Transformer architectures show that Robust CurveMoE consistently improves clean, norm-specific, and Union accuracy over MSD and ERMC. In particular, it improves Union accuracy by 2.37 and 2.13 percentage points over the strongest baseline on CIFAR-100 and ImageNet-100, respectively. Extensive ablations further validate the effectiveness of partial updating, selective expertization, and robustness-constrained expert selection.

Answer accuracy is an insufficient reliability signal for LLM data agents. In structured-data tasks, a benchmark-correct answer can be produced by an invalid trace. This paper introduces Trace Integrity, a deployment reliability criterion for evaluating whether the computation recorded behind an answer is explicit, executable, schema-valid, operator-faithful, replayable, answer-consistent, and auditable. We identify the Structure Gap as the deployment failure mode that makes Trace Integrity necessary: natural-language reasoning and free-form rationales do not reliably specify the operator-level programs required by real-world systems. We operationalize Trace Integrity with execution contracts, structured artifacts that bind user intent to schema elements, operator plans, assumptions, executable queries, verification status, and final-answer linkage. We also introduce CAIT (Correct Answer / Invalid Trace) Rate, which measures how often answer-only evaluation counts computationally unsupported outputs as successes. In an empirical demonstration on BIRD Mini-Dev, Direct SQL, Operation Summary + SQL, and Contract-First SQL achieve answer accuracies of 20%, 22%, and 24%, while their Trace Integrity Pass Rates are 39%, 43%, and 40% and their CAIT Rates remain high at 55%, 59.1%, and 45.8%, showing that answer accuracy, trace validity, and silent-failure risk are distinct evaluation signals. Real-world LLM data agents should, therefore, be evaluated not only by whether their outputs match a reference answer, but by whether those outputs are backed by auditable computation.

When a speaker refers to a scene that the listener cannot directly see, the listener must decide whether to preserve its current understanding or revise it as new utterances arrive. Many language systems treat local mismatch as a cue for updating: divergence from the current understanding encourages adjustment. Yet conversational understanding may be more conservative, interpreting mismatching evidence relative to prior understanding rather than immediately revising it. We introduce a controlled, data-driven framework for turn-by-turn preserve/revise decisions in dialogue, where competing revision policies are learned under otherwise identical conditions. We compare four theory-driven revision strategies, each reflecting a different assumption about when listeners should preserve or revise. Two findings stand out. First, a mismatch-driven policy that updates solely based on local divergence reacts strongly to mismatch but destabilizes grounding and degrades retrieval. Second, an uncertainty-sensitive policy extends mismatch-based updating with accumulated evidence, preserving coherent understanding while maintaining strong retrieval performance. Surprisingly, coherent understanding emerges from a counterintuitive pattern: local mismatch promotes preservation, whereas accumulated uncertainty promotes revision, suggesting that listeners maintain prior understanding despite local mismatch and revise only when uncertainty sufficiently accumulates. This pattern is consistent with conceptual pact theory.

Reliable code retrieval is crucial for developer productivity and effective code reuse. However, current neural code language models (CLMs) powering search tools are susceptible to adversarial attacks targeting non-functional textual elements. In this paper, we introduce a programming language-agnostic, transferable, adversarial attack that exploits this CLM vulnerability. Our approach perturbs identifiers within a code snippet without altering the snippet's functionality to artificially align the code with a target query. We demonstrate that our attack, even when computed using smaller code embedding models, such as CodeT5+, is highly effective and transferable to larger, closed-source embedding models, like Voyage-code-3, or LLMs like Gemini-3.1-Pro. Our attack can increase the similarity between the query and arbitrary, irrelevant code snippets, consequently degrading key retrieval metrics such as the Mean Reciprocal Rank (MRR) of state-of-the-art models by up to 77%. The experimental results highlight the fragility of current code search methods and underscore the need for more robust, semantic-aware approaches.

On-policy self-distillation (OPSD) uses a privileged copy of the student model to provide dense supervision without an external teacher. OPSD keeps this privileged teacher fixed, even though the student distribution and output style change during training. We propose DualOPSD, an asymmetric alternating framework that adapts both policies. The student first learns from the privileged teacher. The teacher then moves toward the updated student distribution on the same student trajectory. This update makes later supervision responsive to the learner and does not require another rollout. On Qwen3-8B in non-thinking mode, DualOPSD improves avg@12 over OPSD by 23.61, 13.89, and 10.00 points on AIME 2024, AIME 2025, and HMMT 2025. Results at 1.7B and 4B show that the accuracy gain depends on model scale. Across all three scales, DualOPSD reduces truncation. The 4B diagnostic also shows lower KL in both directions between the teacher and student.

Multimodal instruction-following models require training data that is accurate, diverse, verifiable, and challenging. Existing synthesis pipelines typically follow a one-pass generate-and-filter paradigm, discarding feedback from failed samples, verifier outcomes, and target-model errors. We present VISA (Visual Instruction Synthesis Agent), an agentic framework that reformulates multimodal instruction synthesis as a self-evolving loop. At each round, VISA analyzes an image to filter incompatible constraints and discover new verifiable ones, samples diversity- and difficulty-aware constraint sets from persistent memory, generates candidate instructions, and verifies the resulting samples with executable tools and structured large language model judges. Failed samples trigger diagnostic-guided recovery, while accepted samples are probed against the target model to estimate difficulty. The resulting verifier signals and target-model failure profiles are written back to memory, allowing subsequent rounds to adaptively expand the constraint space, reduce template repetition, and focus on unresolved model weaknesses. The same verifier contracts further provide reward signals for reinforcement learning without a separately trained reward model. Experiments on MM-IFEval show that VISA consistently improves multimodal instruction following over strong baselines, while preserving general multimodal capability across seven public benchmarks.

An automated theorem prover builds a proof step by step, choosing at each point what to add and what to remove. We cast this construction as a policy acting in a transition system induced by a formal calculus, which fixes which steps are sound: for clausal connection tableaux, leanCoP-style search and plCoP/rlCoP-style planning then become stateful policies over one interface, and policy-learning methods apply directly. We equip such policies with a graph neural network that scores proof edits from structure that transfers across problems, train it by imitation learning from found proofs, and measure how performance holds as we remove search scaffolding, from full symbolic backtracking to a policy the network drives alone. Within a fixed step budget on M2k, MPTP2078-bushy, and TPTP v9.2.1, learned policies solve up to 46% more problems than leanCoP, and reach proofs in an order of magnitude fewer steps.

Large language models (LLMs) remain vulnerable to jailbreak attacks that exploit techniques such as role-playing, obfuscation, code transformation, and multi-step indirection to elicit harmful outputs. As jailbreak strategies keep emerging, defenses have proliferated in an ongoing cat-and-mouse game, yet most remain static: their safety behavior is fixed at deployment, so they cannot accumulate defensive experience or adapt to unseen strategies. We propose a self-evolving test-time defense built around a persistent, cross-interaction rule memory: when an attack succeeds, the framework abstracts that failure into a method-level rule capturing the structural attack wrapper rather than the harmful topic, and reuses it against future inputs. Because rules are method-level, one induced rule generalizes across an entire attack family, and the label space expands as novel wrappers appear. The mechanism operates entirely through external memory and prompting, with no parameter updates, and applies to both open-weight and black-box API models. We realize it as four cooperating modules, but the contribution is the memory-based adaptation mechanism, not the module decomposition. Across four black-box jailbreak families and multiple models, our method substantially reduces attack success rates while preserving benign utility, remains robust under an adaptive composite-wrapper attack, and does not increase over-refusal as the memory grows.

Conversational systems, such as duplex speech language models (SLMs), still lack a streaming, accurate, and empathetic memory system as their soul. We introduce VoiceMem, a simple memory architecture with a parallel informational left brain, an emotional right brain, and streaming memory I/O mechanisms. We further build a complete pipeline for memory-aware SLM training, long-horizon evaluation, and decoupled deployment with interchangeable memory backends. Experiments and real-world deployment show three advantages: i) Accuracy: under top-5 retrieval, the left brain outperforms classical systems such as Mem0 at top-200 by nearly 30 points; ii) Emotional & Personal: the right brain, with short- and long-horizon affective attribution and dual-node persona modeling, achieves state-of-the-art performance across three persona benchmarks and improves the aggregate score by 4.29 points over the previous best system; and iii) Real-Time & Cheap: VoiceMem completes retrieval in 134 ms, well within standard VAD latency, adding no extra conversational delay while maintaining high accuracy and low cost. These results show that VoiceMem provides a practical memory foundation for real-time, personalized, and emotionally aware speech interaction.

Agentic LLM pipelines face escalating inference costs as context accumulates across retrieval, tool use, and multi-turn interactions. To control latency, deployments routinely compress inputs, but this degrades task accuracy. Speculative decoding (SD) accelerates generation losslessly, yet it assumes the drafter and verifier share an identical context, preventing SD from resolving the accuracy-overhead trade-off. We propose AsymSpec, an asymmetric speculative decoding framework that breaks this symmetry: a lightweight drafter reads the full input while the large verifier operates on the compressed view. The drafter steers the verifier via a contrastive $δ$-fusion of logits, modulated by a divergence-aware acceptance gate that preserves verification stability and high draft acceptance rates. Evaluated across four agentic capabilities and two end-to-end agent benchmarks, AsymSpec reaches $\approx 90\%$ of full-context accuracy on average, delivering $1.3$--$1.7\times$ throughput speedups at $0.2$--$0.3\times$ the compute cost on isolated text capabilities. These results show that asymmetric context access yields substantial gains precisely when compression discards critical reasoning signals.

Wearable electrocardiogram (ECG) and photoplethysmogram (PPG) sensors are complementary but individually fragile: motion artifact, poor contact, and sensor dropout can degrade one or both signals. Fusion strategies that assume both modalities are equally trustworthy can become less reliable than a single clean modality under degradation. We present CardioFusion-AI, a framework whose signal-processing front end, including R-peak and systolic-peak detection, an Orphanidou-type signal-quality index, and beat-by-beat pulse transit time estimation, is validated on 53 real intensive-care recordings (848 windows; heart-rate mean absolute error 1.61 bpm for ECG and 2.78 bpm for PPG) and a real annotated fetal ECG database (R-peak F1 0.89-0.98). We then conduct a controlled synthetic degradation study comparing eight ECG-PPG fusion strategies across six degradation regimes spanning graded corruption and complete modality loss, using five independent training seeds. Attention fusion achieved the lowest descriptive overall error (1.66+/-0.43 bpm). Both adaptive gates reallocated weight toward the healthy modality under complete modality loss, but showed near-zero correlation between gate weight and signal quality under graded degradation (r = 0.10-0.24). Signal-quality conditioning produced a specific improvement under missing-PPG conditions (1.56+/-0.59 bpm), approaching the 1.48 bpm unimodal ceiling. With only five training seeds, no pairwise comparison survives Holm-corrected significance testing; effect sizes and confidence intervals are therefore reported. These results indicate that modality availability and modality quality are functionally distinct problems for adaptive fusion.

Linguistic meaning is grounded in conceptual content, from which reference to particular entities emerges as words enter discourse. To examine the processing dynamics associated with these two dimensions of meaning, we selectively disrupted conceptual or referential information in short narratives and traced the resulting effects in human self-paced reading and in the predictive and representational processing of large language models. In human reading, conceptual disruptions produced a strong but localized processing cost, emerging immediately after the distorted word, reaching an early maximum, and then declining rapidly. Referential disruptions produced weaker effects, which decreased more gradually across subsequent words, and were more strongly modulated by sentence boundaries. In the language model, both disruptions emerged immediately at the manipulated word. Contextual model surprisal showed a pattern closely paralleling human reading: conceptual disruption produced a larger, more locally concentrated effect that decayed rapidly, whereas referential disruption produced a smaller and more gradual downstream effect. Output-layer representations showed a different pattern: referential disruption produced a larger initial displacement, while both distortions were subsequently characterized by power-law decay. Together, these results provide convergent evidence for distinguishable processing dynamics of two types of meaning: conceptual information imposes a more locally concentrated integration cost, whereas referential information engages a more distributed process of maintaining discourse-level identity.

Multi-agent large language model (LLM) workflows have emerged as a powerful paradigm for solving complex, open-ended tasks through collaborative reasoning among specialized LLM agents, but they incur substantial operating costs due to repeated LLM invocations and long-horizon context accumulation. Existing cascade routing methods make one-shot, query-level decisions and cannot adapt to the dynamic, state-dependent nature of multi-step workflows, in which the right LLM at each step depends on evolving task progress, remaining task difficulty, and cost-efficiency requirements. We present ProgRouter, an online progress-guided routing framework that adaptively selects LLM agents across workflow steps to preserve task-solving quality while adhering to time and cost budgets. ProgRouter introduces a multi-view task progress scorer that combines coarse workflow outcome regimes with fine-grained signals on subtask completion, progress trends, and workflow state quality. Then, a dual-path task progress predictor and an adaptive meta-gating mechanism estimate the progress gain for each candidate routed LLM. ProgRouter makes online step-wise routing decisions that balance progress gain, task time budgets, and long-term operating cost efficiency. Experiments on HumanEval Plus, MBPP, MATH-500, and ASQA, spanning agentic code generation, mathematical reasoning, and retrieval-augmented long-form question answering, demonstrate that ProgRouter reduces the operating cost relative to key baselines while maintaining strong task-solving performance.

Orthogonal optimisers such as Muon can substantially accelerate large language model pretraining relative to Adam, yet the mechanism remains incompletely understood. We investigate this through an out-of-sample spectral probing analysis of Transformer loss landscapes. At checkpoints along real training trajectories, we decompose each momentum buffer into its singular directions and estimate the loss-optimal step size along each direction on held-out data. The resulting spectral profile is anisotropic yet stable across batches and training stages, and consistent across the optimisers and model scales: a volatile head operating at the Edge-of-Stability supports a much smaller step size than the tolerant bulk, which permits substantially larger steps. This profile provides a unified spectral allocation account of why Muon outperforms Adam, which outperforms SGD. It also exposes a limitation of Muon's uniform scaling: it still underutilises the bulk. Guided by this finding, we introduce Spectral-Aware Muon (SAMuon), which holds the head at the Muon scale and amplifies the bulk using a static spectral prior. We provide two variants: the complete SAMuon follows the measured profile using a low-rank randomised SVD and the simplified SAMuon-lite uses a two-level approximation via rank-one power iteration. Neither method adds persistent optimiser state or notable extra FLOPs beyond Muon at scale, and the idealised exact-whitening versions of both retain Muon's asymptotic convergence rate under standard assumptions. Across "modded-nanogpt" models from 124M to 1B parameters, both variants outperform tuned AdamW and Muon (Scion implementation) baselines in all evaluated model-scale and batch-size configurations. SAMuon requires 13.3% to 24.0% fewer training tokens to reach the same validation loss as Muon, while SAMuon-lite retains most of this gain with near-zero wall-clock overhead.

Retrieval-augmented generation (RAG) is widely used to mitigate hallucination issues in large language models (LLMs) and multimodal large language models (MLLMs). In particular, knowledge graph (KG)-based RAG leverages structured knowledge to provide (M)LLMs with high-quality external information. Building on these works, recent studies have explored multimodal knowledge graphs (MMKGs) as knowledge bases for GraphRAG. This enables Graph RAG to integrate knowledge across multiple modalities, thereby further enhancing its performance. However, existing MMKG-based RAG methods generally follow a common pipeline in which different modalities are largely processed independently before being fusion. As a result, textual context is only used to a limited extent during visual information extraction and subsequent multimodal knowledge fusion. This brings a semantic gap between images and text which limits the multimodal GraphRAG performance. To address this issue, we propose a novel framework for constructing a Context-Enhanced MMKG (CEMMKG) to better support multimodal GraphRAG. The proposed CEMMKG enriches each image with complementary textual context at both local and global scopes. Local context goes beyond the surrounding text by incorporating sentences that are semantically related to the image, while global context provides a summary of the entire passage. We further introduce a multi-granularity design for the local context, allowing it to capture semantically relevant information at different levels of detail. Extensive experiments on the selected vision-centric dataset validate that CEMMKG is effective in leveraging contextual information to improve MMKG-based RAG performance. Moreover, its effectiveness across different MMKG-based RAG methods demonstrates its broad applicability.

Subgroup performance differences are the standard evidence for fairness bias in medical imaging, and the usual response removes the demographic information that a model encodes. Here we introduce Fair-model Reference And Mechanism Evaluation (FRAME), a two-step framework for auditing such a claim. The first step derives a fair-model reference, the distribution of the difference under exact fairness at the observed subgroup sizes. In the second step, we test the remainder with two operators in representation space. One operator cannot change a within-group ranking by construction. Across 702,206 images and 36 encoders, the reference accounts for a median 41% of the reported race difference and 22% of the age difference. Injecting demographic decodability leaves the remainder unchanged, while entangling the group with the disease direction raises the race difference from 0.077 to 0.118. No intervention we tested changes the remainder more than a change of random seed does. Those interventions reduce a difference at the operating point and leave the within-group ranking difference at a median of 0.000. Applied to 89 differences in 9 published studies across 6 medical imaging modalities, the reference accounts for a median 25% of a rate difference and 70% of a difference in the area under the receiver operating characteristic curve. Image-text pretraining instead raises worst-group performance by about 0.05. Applying FRAME before choosing an intervention could distinguish differences that need a mechanistic explanation from differences compatible with sampling variation at the current cohort sizes.

Personality is increasingly important in large language models (LLMs), as it shapes users' trust, engagement, and emotional experiences. While the Myers--Briggs Type Indicator (MBTI) has emerged as a common framework for assessing LLMs' personality, existing studies focus primarily on full-precision models and evaluate only final outputs. They overlook the widespread deployment of quantized LLMs requiring low memory footprints, whose personality traits remain underexplored. In this work, we present a systematic MBTI analysis of open-source LLMs across multiple precisions, including mainstream 4-bit methods (GPTQ, AWQ) and extreme 2-bit settings (AQLM variants). Beyond output-level evaluation, we examine how personality emerges across layers through option-level entropy and confidence-gap dynamics, and introduce Uncertainty-Amplified Layer Decoding (UALD) to study decoding-induced personality drift at inference time. Our results reveal a key insight: LLMs' personality is not a static property, but an emergent, layer-dependent decision process sensitive to quantization, prompting, and decoding. Specifically, we find that (1) ENFJ remains dominant across model families and precisions; (2) 4-bit quantization largely preserves coarse personality structure, while 2-bit quantization disrupts fine-grained prompt consistency and cross-precision agreement; (3) personality decisions emerges in upper layers, following substantial ambiguity in early layers; and (4) inference decoding can shift personality, while personality-aligned conditioning improves robustness. These findings provide a new perspective on the behavioral reliability of quantized LLMs and highlight the importance of considering internal dynamics and inference strategies in personality-sensitive chatbot applications.

International adoptees retain phonological traces of a birth language they can no longer speak or comprehend, a persistence typically attributed to a biologically-timed critical period. We asked whether it could instead reflect the ordinary dynamics of learning, using automatic speech recognition models that simulate the international adoptee experience without maturational confounds. Models were trained on one language and then abruptly switched to a second. We found that traces of the first language persisted throughout second-language training, but mainly in the lowest, pre-phonemic layers. These traces were functional, as models with early exposure re-learned their lost first language 14% faster than naive models; this advantage held even against models adopted early from a related language and disappeared when the earliest layers were substituted from a non-adopted model. We argue that these critical-period effects reflect entrenchment of foundational representations rather than a maturational loss of plasticity, and that experience plays a central role in critical periods in language acquisition.

Understanding instruction-following capabilities in scientific domains is essential for effectively leveraging Multimodal Large Language Models (MLLMs) to advance the development of scientific fields. In this work, we introduce SciMIF, a novel benchmark designed to evaluate the capability of MLLMs in following complex scientific instructions. Specifically, based on an extensive analysis of 22 distinct tasks across 5 representative scientific disciplines, we propose a comprehensive taxonomy comprising 10 constraint groups that captures both general functional requirements and discipline-specific characteristics. Guided by this taxonomy, we develop a high-fidelity instruction injection pipeline to systematically augment existing scientific datasets. We conduct comprehensive experiments on multiple state-of-the-art closed-source and open-source MLLMs. Our findings reveal significant performance disparities across different scientific disciplines, with chemistry posing greater challenges for current MLLMs. Furthermore, we observe that increasing the model scale does not yield corresponding improvements in constraint adherence, and current models still struggle severely with fine-grained constraints and instructions requiring the deep application of disciplinary knowledge. SciMIF fills the current void in evaluating multimodal instruction adherence within scientific domains, laying a crucial foundation for future enhancements of MLLMs in rigorous scientific applications. Data and code will be released at https://github.com/shenye7436/SciMIF .

Multimodal medical prediction often faces incomplete pairing: auxiliary modalities with complementary signal are available for only a subset of subjects (or none) and cannot be assumed at deployment. We introduce PANDA (Prototype Anchored Data Alignment), a two-stage framework that transfers auxiliary information to a primary-modality model without auxiliary inputs at inference. Stage 1 learns a shared embedding from the paired subset and estimates class prototypes from auxiliary modalities; Stage 2 trains the primary encoder on all subjects using cross-entropy plus alignment to the frozen prototypes. Because supervision is defined at the class-prototype level, PANDA accommodates arbitrary pairing rates, including zero subject overlap. We evaluate PANDA on two applications. On a 1,021-subject multi-scanner ADNI cohort, we perform AD/CN classification with three auxiliary modalities at distinct pairing rates: tabular scores (44.8%), FDG-PET (18.7%), and external handwriting kinematics (0% overlap). Relative to the same-backbone MRI-only baseline, PANDA attains AUC 0.868 +-0.020 (+7.9pp) and reduces 1.5T CN false positives by 24.3pp; on a fully trainable Conv5-FC3 backbone it reaches AUC 0.893 (best overall). A pairing-rate ablation shows that the joint anchor remains within seed noise from 75% to 5% pairing. On TCGA-Lung survival prediction from whole-slide images with RNA-seq as auxiliary data, PANDA improves over WSI-only on 2-year OS (AUC +3.5pp) and Cox PH (C-index +9.0pts) and outperforms full-fusion training, which underperforms WSI-only, while requiring no RNA at inference; wide confidence intervals on this smaller cohort keep the gains below conventional significance. Overall, PANDA provides a deployment-oriented mechanism for leveraging incomplete auxiliary modalities to improve primary-modality prediction.

Graph-based RAG improves multi-hop question answering by organizing evidence as a knowledge graph. However, most existing RAG systems process each query in isolation and discard useful reasoning from the LLM's response after inference. As a result, later related queries need to retrieve evidence and reason from scratch. We propose LivingRAG, a Graph RAG framework with writable and reusable reasoning experience. LivingRAG adds a writable experience store to a graph-based retrieval backbone, enabling verified experiences to be reused during inference in two ways. Stored graph signals help retrieval find entities and passages that were useful in earlier related queries. Stored summaries provide a reference reasoning pattern for answer generation. We analyze online QA streams and find reusable signals from shared entities, graph neighborhoods, and question templates. Experiments on multi-hop QA benchmarks show that LivingRAG improves accuracy over strong RAG baselines and reduces completion-token use when relevant prior experience is reused.

Autonomous R\&D agents now write, run, and improve executable artifacts under automated evaluation---but largely as laboratory instruments: shown on curated benchmarks, with gains that are hard to trace to a cause and costs well above what sustained engineering practice absorbs. The limitation is structural. Most systems treat each attempt as nearly self-contained, so logs, memories, and search trees record what happened without establishing which design element produced an improvement, whether its evidence survived validation, or how it recombines with others. Long campaigns therefore keep re-learning the same lessons. We introduce Praxist, a lineage-centered generational system that converts reproducible artifacts and evaluator outcomes into a typed evidence graph of findings, lane-structured frontiers, and agendas. Separating local artifact construction from cohort-level evidence synthesis lets later attempts inherit validated mechanisms, unresolved claims, and useful constraints, and leaves results attached to an inspectable lineage. On the standardized 75-task MLE-bench suite, the finalized official-grader results give Praxist 60 medals (80.0\%), 49 of them gold, against 55 medals (73.3\%) and 34 gold for a Claude Code baseline on Claude Opus 4.8---at a recorded model spend of US\$3,054 versus US\$38,370, roughly a twelfth of the cost. Four case studies---quantitative trading, LiDAR-inertial-visual SLAM, tokamak magnetic control, and rocket landing---carry the same process into open-ended engineering problems, improving on each task-native baseline in headline accuracy, survival, or resource cost, with the discovery path on record. Stronger artifacts at an order of magnitude less spend, each backed by an auditable lineage, are, to our knowledge, first brought together here: the operating profile production research requires, not the one a benchmark demonstration establishes.

The rapid advancement of Large Language Models (LLMs) makes it increasingly difficult to distinguish human writing from machine-generated text. Training-free detection offers a scalable solution, yet common confidence-based metrics mainly measure average token probabilities and often miss the signal fluctuations that characterize human writing, which we call "generative vitality". Spectral analysis offers a way to capture this vitality, but its mechanism and practical boundaries remain underexplored. In this paper, we analyze spectral detection from both theoretical and empirical perspectives. We connect spectral energy to variance in proxy log-probability trajectories and explain how broader human token choices create the fluctuations used by frequency-domain indicators. We further show that the strength of this signal depends on text length and sampling range: spectral evidence is clearest for long, continuous, constrained generation, while short, fragmented, mixed, and edited settings require complementary confidence and fluctuation views. These findings clarify when frequency-domain detection works and provide guidance for future multi-dimensional detector design.

Sparse autoencoders (SAEs) are widely used to interpret the internal representations of large language models (LLMs), yet their reliability under post-hoc model compression remains poorly understood. We present a systematic study of how pruning affects SAE behavior and theoretically show that, for a fixed SAE, its impact is governed by perturbation energy, a covariance-weighted norm. This perspective exposes a key limitation of magnitude pruning: by ignoring activation geometry, it distorts the learned representation space and degrades SAE functionality. Activation-aware methods such as Wanda and SparseGPT, in contrast, implicitly control perturbation energy and are therefore substantially more robust at preserving SAE behavior. We further reveal a consistent structural vulnerability across all pruning methods: middle layers are significantly more sensitive to pruning than early or late layers. Guided by this insight, we propose a layer-wise sparsity allocation strategy, achieving lower perplexity under the same average pruning sparsity. Experiments across four model architectures validate our theoretical findings. Code is publicly available at https://github.com/osu-srml/sae-robustness-under-pruning/tree/main.

Physical AI models are evaluated on suites of benchmarks that differ across model reports, leaving the model-by-benchmark matrix sparse and the relationship between benchmarks unmeasured. We construct a matrix of 51 models on 12 physical AI benchmarks, selected from a registry of 51 benchmarks and 152 models by reporting density, combining scores from model cards and benchmark papers with our own evaluation runs under each benchmark's official protocol. We measure how much information the benchmarks share and show quantitative evidence of Redundancy. Redundancy affects reported rankings: collapsing the two substitute pairs into single columns moves 22 of 51 models by three or more places under an equally weighted average. We then select benchmarks greedily under a utility combining score dispersion with variance not explained by the already-selected set, and obtain a four-benchmark subset retaining 78.5\% of the utility of all 12, on which we fit a Bradley--Terry ranking. The procedure requires only benchmark-level scores with sufficient overlap and is not specific to physical AI.

Large language models (LLMs) have shown promise for automated unit test generation, but existing evaluations largely rely on standalone settings and a narrow set of programming languages, overestimating real-world readiness. We introduce XREPOTEST, a multilingual repository-level benchmark for unit test generation spanning five underexplored languages: Rust, Go, Julia, PHP, and Ruby. XREPOTEST evaluates tests under realistic repository constraints using a containerized execution framework and multiple context augmentation strategies, including file-level, LSP-based, and retrieval-based context. Beyond standard metrics such as test pass rate and coverage, we propose Invocation Rate (IR) to assess whether generated tests meaningfully exercise the intended functionality. Experiments with 14 state-of-the-art LLMs, including Claude 4.5, GPT-5.2, DeepSeek V4-Pro, and Qwen families, reveal a substantial gap between standalone and repository-level performance, as well as trade-offs between richer context and test reliability. Overall, XREPOTEST provides a challenging and informative benchmark to advance scalable and robust unit test generation in realistic software environments. The dataset and code are publicly available at: https://github.com/solis-team/XRepoTest

Multi-agent systems (MAS) sometimes already have the potential to answer correctly, but still report a wrong answer. Explaining this outcome is difficult because generation, communication and final answer-selection rules usually change simultaneously. We conceptualize multi-agent reasoning as an evolutionary pipeline of candidate generation, peer communication and terminal selection, wherein consensus without quality control can exhibit patterns of memetic drift. We study two questions: (1) when an LLM judge provides effective selection pressure by supplying a signal of answer correctness for candidates generated in a multi-agent system, and (2) when using that signal improves the reported answer. To map judge reliability, we analysed 15,336 questions from MMLU-Pro, GPQA, MedXpertQA and MuSR, with Humanity's Last Exam analysed separately. To test these rules, we replayed 81,390 fixed candidate pools drawn from 16,278 questions across five benchmarks. We report three findings. (1) A correct answer is often already present among the generated candidates, but the system can still converge on and report a wrong answer. (2) Judge reliability is not a fixed trait of the model, but varies with the task, the generator and how rare the correct answer is. (3) Combining answer frequency with the judge's evaluation changed only the final answer-selection rule and raised accuracy from 63.82% to 70.82-70.95%, primarily by rescuing correct answers that were outnumbered by popular errors. In the systems studied here, the value of generating more candidates depends on whether those extra samples make correct answers present, frequent or recognisable. By isolating generation, recognition and selection, these findings establish a diagnostic basis for designing multi-agent architectures that protect generated correct answers from being lost.

On-policy distillation trains a language model on its own generations while a teacher scores them token by token. It combines the dense supervision of imitation learning with the on-policy sampling of reinforcement learning. But it requires a second, larger model to act as teacher. On-Policy Self-Distillation (OPSD) removes that cost. The teacher is the model itself, conditioned on privileged information the student will not have at test time, such as a reference solution, a plan, or environment feedback. The teacher is no stronger than the student, only better informed. Early results were promising, with accuracy comparable to reinforcement learning at a fraction of the generated tokens. But the same asymmetry that produces the signal also biases it. One failure mode now dominates the field: collapse, the progressive narrowing of the set of reasoning paths the model can produce. Collapse is not specific to OPSD, though privileged information aggravates it. This review treats collapse as a symptom governed by three levers: (i) where the signal is applied, that is, how tokens are weighted; (ii) what the teacher is shown, that is, the nature of the privileged information; and (iii) when the signal changes, that is, the teacher's dynamics and the decay of guidance. We restrict our scope to mathematical reasoning, where the method originated and where its failure modes are best documented. We report no new experiments. The contribution is structural: a shared vocabulary for phenomena named differently across papers, and a clear line between what is settled and what is still disputed.

Traffic Anomaly Understanding (TAU) requires models and systems to detect, reason about, and explain anomalous events in transportation videos. To address this challenge, we propose TAU-Agent, an agentic retrieval-augmented framework for traffic anomaly understanding. Given a task query, a central retrieval agent orchestrates two visual perception tools, namely a Video Captioning Tool and an Open-Vocabulary Tracking Tool, to retrieve and select query-relevant evidence, including captions, temporal intervals, and object trajectories. The selected evidence, together with sampled video frames and the input query, is provided to a supervised fine-tuned vision-language model for final reasoning and answer generation. We evaluate TAU-Agent on both the in-domain and the out-of-domain benchmarks from the AI City Challenge 2026. TAU-Agent achieves scores of 0.6779 on Track 3, 0.3998 on Track 7, and 67.9275 on Track 8, ranking second, twelfth, and fifth, respectively. Code is available at: https://github.com/siri-rouser/TAU-Agent.

Automated fact-checking (AFC) systems retrieve evidence and predict claim veracity, yet evaluations omit simple baselines, systems are developed for a single benchmark and cannot be trusted to generalise across domains. No prior work cross-evaluates the full two-stage retrieve-then-verify pipeline across diverse datasets, complementing retrieval-only studies (Thakur et al., 2021) and single-stage benchmarking studies (Calamai et al., 2025). We benchmark nine models, ranging from random and sparse baselines to fine-tuned transformers, zero-shot LLMs, and the two highest-ranked systems from the AVeriTeC 2025 shared task, across four datasets spanning scientific, open-web, and climate domains. Three findings stand out: (1) on ClimateCheck claim-only and fine-tuned models outperform zero-shot LLM and top-performing AVeriTeC 2025 systems, highlighting that noisy evidence can degrade veracity prediction; (2) system rankings are strongly domain- and metric-dependent: the best model on SciFact (macro-F1 0.70) drops to 0.31 on ClimateCheck, while the AVeriTeC 2025 winner and runner-up swap rankings based on evaluation metrics and datasets; (3) replacing retrieved evidence with gold annotations improves veracity accuracy by 14-22 points across models, confirming retrieval remains primary bottleneck. We release code, pre-processed datasets, and all results to support reproducible AFC research.

*Chulin Zhao and Ruoqi Hu contributed equally to this work. State-of-the-art text-to-image (T2I) models exhibit pronounced and systematic defects when prompts involve intricate compositional factors such as multiple entities and multiple attributes. In this paper, we investigate how humans identify such defects. Specifically, we manually select 651 reference images from the four categories of people, hand, object, and scene that exhibit complex compositional characteristics, from which prompts emphasizing compositional factors are derived by manually editing ChatGPT-generated prompts. We then feed the prompts into three selected T2I models to generate AI images and conduct a comprehensive subjective study to identify their defects. For each image, 29 participants provide multi-label assessments specifying defect types and locations. The study yields the compositional AI-generated image defect (CO-AID) dataset, including reference images, prompts, AI-generated images, and information on defect locations and types. Experimental results show that training a deep model on CO-AID can both predict defects in AI-generated images and optimize AI image generation, demonstrating its usability and effectiveness. The database and supplementary materials are available at: https://github.com/Future-IQA/CO-AID .

Three-dimensional probabilistic inversion of time-domain airborne electromagnetic (AEM) data is limited by the cost of the forward solve. Even though one simulation takes only tens of seconds, a Bayesian inversion of a survey of millions of soundings requires of order $10^{10}$ forward evaluations. To address this, we develop a continually learning neural-operator surrogate of the three-dimensional AEM forward operator that replaces the solver inside the Bayesian inversion. We start from the point of view that regardless of what geological prior is specified, Maxwell's laws remain invariant. Secondly, we avoid the limitation of learning on a single prior by continual learning on consecutive priors, which means our surrogate becomes richer as it is applied in future case studies, either by the authors, or by the scientific community. We use a validity check built on ensemble disagreement to divert cases with measurements outside the training range to the solver. Driven by the surrogate, the identical Markov chain Monte Carlo sampler reproduces the full-solver posterior, and its credible intervals cover the truth within 2.6 percentage points. Applied to the 2013 Capricorn TEMPEST survey in Western Australia, the surrogate inverts over two million soundings in seconds, a computation infeasible for the solver. Testing the geological prior against the entire survey costs minutes. The framework delivers uncertainty-quantified conductivity imaging at survey scale, which we believe is essential to perform near real-time mineral-systems targeting with geophysics.

Control variables are widely used in statistical modelling to account for omitted variable bias of known confounders. However, they have largely been underexplored in deep learning. This is surprising, given that deep learning models encode image-inferable covariates, such as demographic variables, into their predictions when these covariates are correlated with the outcome---a form of omitted variable bias referred to as 'shortcut learning'. While many existing confound-control or fairness methods try to restrict the correlation of such covariates with model predictions, we show that this fails to correct for omitted variable bias. We therefore propose a control variable approach for deep learning models, based on generalised additive modelling of the effects of model inputs and covariates. As flexible additive models can suffer from concurvity, we introduce an estimation procedure that refits the final layer of a pre-trained network to include covariate effects, using cross-fitting with ridge penalisation. We show how these effects can be orthogonalised with respect to covariates to exclude their mediated effects and that model predictions can be marginalised over the covariate distribution to control for their effect. This yields unbiased, interpretable predictions and offers flexibility to model the desired effects depending on the scientific or fairness objective. We verify our approach using simulated images, and demonstrate consistent estimation of true effects. Existing methods either require more data or fail to recover the true effects. We apply our method to real neuroimaging data with experimentally induced confounding, where it recovers prediction performance to near the level of a model trained on unconfounded data. Code is available at https://github.com/mpff/cocodeel.

World models aim to simulate how complex environments evolve under actions and events, yet existing video-based world models primarily learn dynamics from visual observations, which reveal outcomes rather than the underlying knowledge, rules, and mechanisms governing world evolution. This makes it difficult to maintain persistent consequences and support coherent, open-ended evolution. We introduce Code World Model, a framework that separates world evolution from visual realization by combining the reasoning and coding capabilities of language models with the generative priors of video models. A coding agent serves as the world brain, reasoning about events and their consequences and generating executable code to maintain persistent world state and perform rule-consistent evolution. To connect executable state with visual generation, we introduce a proxy representation that encodes frame-wise spatiotemporal constraints and is compiled into a proxy video, which conditions a video model to render high-fidelity visual observations. We further develop data pipelines for constructing aligned proxy-observation pairs from gameplay and real-world videos. After fine-tuning on paired gameplay data, MiniMax-H3 follows proxy-based spatiotemporal specifications from simple interactive worlds built by the coding agent while preserving rich visual details and dynamics. These results demonstrate the potential of combining code for persistent world evolution with video models for flexible visual realization, providing a new path toward open-ended world models.

Air traffic control procedures are executed through spoken exchanges between controllers and pilots. These interactions are essential to the safety of air transportation: failures in their execution can create severe operational hazards, as evidenced by past fatal accidents. Assessing whether an instruction has been followed requires relating what was said to the aircraft concerned, its state, and the obligations that pilots must meet. We present a runtime verification framework that monitors such procedures by checking controller-pilot exchanges, surveillance data, and onboard observations. The framework parses radio communications into events linked to the entities they concern and merges them with surveillance and onboard observations into a time-stamped trace. The ICAO-derived obligations as formalized as temporal formulas with explicit time bounds and evaluated over execution traces. Every violation is reported along with the breached obligations and the observations that support the verdict. With real traffic, the complete pipeline reaches an F1 of 0.85 against blind human-annotated violations; in 1,495 synthetic situations derived from two public corpora, the monitor logic returns the expected verdict in every case. In two historical accidents reconstructed from official investigation reports, the monitor identifies the same procedural deviations documented by the investigators.

GNN-based Knowledge Graph Question Answering (KGQA) pipelines process queries through four discrete stages: entity linking, subgraph retrieval, GNN reasoning, and answer generation. Standard robustness evaluations conflate stage-level failures into a single end-to-end metric, obscuring both the source of brittleness and the appropriate mitigation target. We ask which stage fails, and why, when the pipeline is subjected to adversarial perturbations on the input question. We introduce a stage-isolation protocol with two answer-preserving adversarial perturbations verified against the knowledge graph: Compositional Restructuring (CR) and Relation Synonym Swap (RS) target distinct stages while leaving entity seeds intact. Evaluated across ComplexWebQuestions and WebQSP, the results run counter to prevailing assumptions: the GNN reasoning stage retains near-baseline accuracy when the subgraph is intact, while subgraph construction accounts for over 99\% of the end-to-end collapse under CR, occurring even when the gold answer is present in 74\% of retrieved subgraphs. This exposes a fundamental distinction between answer presence and answer reachability that end-to-end metrics cannot detect, and places the mitigation target firmly at the subgraph construction stage rather than the reasoning model. Perturbed datasets and evaluation infrastructure are released at https://anonymous.4open.science/r/atkgrag-E85C .

As large language model (LLM)-based multi-agent systems (MASs) are increasingly applied to long-horizon complex tasks, their reliability has emerged as the core bottleneck hindering their real-world deployment. Existing MAS debugging and repair methods typically rely on rerunning and resampling the entire execution trajectory. However, a fundamental question remains to be answered: do these methods causally repair MAS failures or merely stochastically repair by leveraging the randomness of LLM sampling? To evaluate the effectiveness of MAS repair methods, we introduce SymTrace, a controlled evaluation framework that records the MAS execution trajectory and establishes intervention anchors. During replay, it effectively reconstructs the execution before the anchor using recorded logs and only regenerates the downstream trajectory, thereby enabling the reliable reproduction of MAS failures. We further construct the dataset SymFail, comprising 536 human-annotated failure trajectories with graph-linked locations, categories, and trace evidence. Based on these foundations, we conduct a large-scale empirical study across three mainstream MAS frameworks. Our findings reveal that existing unguided rerun methods are highly unreliable, exhibiting low failure reproduction and repair rates (only 67.97% and 6.90%, respectively). Building upon these findings, we further explore the effectiveness of a symptom-driven intervention method, which successfully repairs 20.15% of the failed cases (a 191.89% improvement to state-of-the-art repair methods). This study aims to provide actionable insights for MAS debugging and repair research, paving the way for the robust deployment of multi-agent systems.

Game-theoretic models provide principled frameworks for modeling vehicle interactions, but their underlying temporal assumptions have not been systematically examined against real-world driving behavior. In particular, it remains unclear how simultaneous, sequential, and asymmetric interaction structures can be measured from vehicle trajectories. This paper develops a trajectory-based interaction measurement framework to identify interaction events and quantify behavioral change onset, temporal organization, post-onset response dynamics, and ordering stability. The framework uses behavioral deviations to verify candidate interactions. We evaluate the framework on six real-world trajectory datasets, including INTERACTION, highD, inD, rounD, Waymo Open Motion, and nuPlan, covering diverse road geometries, traffic environments, and interaction types. The results show that concurrent and sequential behavioral changes both constitute substantial proportions of observed following, merging, and conflicting interactions. Among sequential interactions, stable ordering is more prevalent than alternating ordering, indicating that persistent asymmetric roles are a common interaction structure. Importantly, temporal precedence does not necessarily coincide with a measurable behavioral response, indicating that temporal ordering alone may not be sufficient to characterize behavioral dependence. These findings show that real-world interactions exhibit concurrent, sequential, and persistently ordered temporal structures. Different game-theoretic formulations are therefore better regarded as complementary modeling abstractions for different interaction regimes rather than as a universal structure governing all vehicle interactions.

Machine Learning (ML) is increasingly used in virtual prototypes of embedded systems to model behaviors that are difficult to capture analytically. However, integrating ML models into virtual platform simulation is still typically done through ad hoc solutions, which limits reuse, comparability, and reproducibility. This paper presents \textbf{\textit{SAMpLE}}, an open-source SystemC-AMS-based framework that integrates ML models as first-class Timed Dataflow (TDF) components through a standardized plug-and-play interface. SAMpLE provides two execution backends: a native C++ backend for online training of lightweight models, and an offline backend for executing externally developed models without requiring re-implementation in C++ or manual integration steps. The framework uses ONNX as a standard model exchange format to enable integration of externally trained ML models into SystemC-AMS simulations, and allows the evaluation of different ML-based solutions within the same testbench, dataset, and simulation workflow. The modular design and unified and reproducible environment will allow future extensions of SAMpLE to new models, without modifying the SystemC-AMS structure.

Driver behavior is heterogeneous, context-dependent, and changes over time, and these properties shape the traffic phenomena we observe. Most models, however, fix in advance which behavioral variables interact and how. Behavior outside that form is absorbed as noise, while models flexible enough to capture it tend to lose interpretability. We introduce a quantum-inspired representation of driver behavior that combines properties usually treated separately or in part: it is continuous, probabilistic, context-dependent, history-dependent, and represents interactions among behavioral variables as learned from data. Each driver is encoded as an evolving density matrix, providing a unified representation of behavioral uncertainty, temporal evolution, and context-dependent behavioral variation. Trained without supervision on the I-24 MOTION dataset, the framework recovers three interpretable driving profiles representing three regimes: free flow, transition, and congestion. The profiles capture the behavioral range of the data and the smooth transitions drivers make between regimes as conditions change. The same representation also reproduces known macroscopic phenomena, aligning with the fundamental diagram and reproducing hysteresis loops. We also show how the representation supports practical use: it supplies context-dependent parameters to classical car-following models, and gives an autonomous vehicle a live behavioral read of the surrounding drivers with a short-horizon forecast of their motion. The framework points toward models of traffic that are interpretable and trustworthy by construction. We release an open-source toolkit on GitHub (https://github.com/mselayan/quantum-driver-representation) spanning data processing, training, inference, and analysis.

Software vulnerability (SV) assessment helps prioritize remediation by characterizing reported vulnerabilities. Existing automated methods predict assessment results from SV reports (SVRs), but often overlook information in rich text, such as screenshots and code snippets, as well as contextual information about vulnerable projects. They also focus on prediction accuracy without providing explanations or supporting evidence, limiting their practical use when analysts must validate imperfect predictions. We propose EAVA, a framework that uses large language models (LLMs) to assess SVs and provide supporting evidence. EAVA employs specialized LLM agents to process rich-text content and project information, and builds a dedicated assessment model through a two-stage training pipeline. It first uses supervised instruction tuning on automatically annotated reasoning trajectories to inject domain knowledge, and then applies reinforcement learning to improve intrinsic reasoning. EAVA also retrieves similar historical vulnerabilities as supplementary evidence. Experiments on a newly collected SVR dataset show that EAVA outperforms the strongest baseline by 5.3 to 35.2 percent across multiple metrics. Ablation studies confirm the effectiveness of assessment-specific model training and information enrichment. A user study with security experts further demonstrates that the evidence provided by EAVA is useful and practical for real-world SV assessment.

Multilingual language models transfer knowledge across languages through shared subword vocabulary, a mechanism that breaks down when related languages use different writing systems. Prior work addresses this via script equalization (romanization or IPA transcription), but direct comparisons are rare; the focus has been on encoder-only models, with most work adapting existing pretrained models. We systematically compare different input representations in autoregressive multilingual pretraining, comparing orthographic text, IPA, and romanization in a controlled setup across three scales (467M, 709M, and 1.03B) on eight languages in four typologically motivated pairs. Across a wide range of downstream tasks on seen and unseen languages, romanized pretraining yields the strongest cross-lingual transfer, and the advantage over text widens with scale. IPA improves over text in most settings but trails romanization. Surprisingly, finetuning a text-pretrained model on romanized data hurts performance on languages already covered by the base model, only marginally helping when the model lacks script coverage. Our results indicate that for multilingual models spanning typologically diverse scripts, to obtain maximum benefits, romanization should be treated as a core design choice applied at pretraining rather than a post hoc fix.

Relational Deep Learning (RDL) is an effective approach to machine learning over multi-table relational databases. In RDL, a database is modeled as a graph in which each row is a node and each foreign-key relation is an edge, and a graph neural network (GNN) is trained on this graph. Training a GNN requires sampling a subgraph around every seed node in the training set, and the cost of training is largely determined by the size of these subgraphs. This paper aims to reduce subgraph size by leveraging the join and aggregation capabilities of relational database systems. We observe that sampled subgraphs are obtained by following metapaths composed of foreign-key links, and that many of these metapaths can be pruned without loss of accuracy. We present MetaSieve, a metapath selection layer that determines which metapaths to retain and which to prune. For each candidate metapath extension, MetaSieve computes statistics via SQL join and aggregation queries and evaluates the extension based on a novel scoring function that prefers lightweight but informative candidates. Metapaths whose scores fall below a threshold are deemed uninformative and pruned. Metapath selection in MetaSieve is lightweight since it relies only on database statistics and task labels, and it is independent of GNN parameters, so it integrates with diverse GNN architectures for classification and regression. Our evaluation on the RelBench benchmark with multiple GNN backbones shows that MetaSieve consistently reduces per-epoch training time by large margins while maintaining and often improving accuracy.

Forecasting a stochastic dynamical system rarely means a single number: one wants several observables---future state, threshold event, regime label---each with its own likelihood. Standard multi-task recipes balance per-task losses, tuned or learned. We instead compose the observables' likelihoods in per-task free-routed last-layer beliefs on a shared backbone; this absorbs unit-dependent loss scaling into likelihood parameters learned in the same gradient pass. Stochastic dynamics supply what static benchmarks cannot: computable ground truth for the predictive variance. Results land where theory puts them: on the well-specified, homoscedastic Ornstein--Uhlenbeck process the learned predictive law recovers the analytic kernel and correctly specified baselines tie. On heteroscedastic systems (stochastic Lorenz-63, real air-quality data) the belief's input-dependent variance separates: best single-run NLL on the state and regime tasks, calibration matched only by arms whose NLL it beats, at a fraction of the tuned grids' cost. On the real series the state margin holds across five rolling origins.

Explaining deep learning models operating on time series data is crucial in various applications that require transparent and interpretable insights into model behavior. {Existing explanation methods generally fall into two categories: attribution-based explanations, which identify the temporal regions most responsible for a prediction, and counterfactual explanations, which reveal how an input should be modified to alter the model's decision.} {Despite valuable insights, these two fields are largely studied independently. This disconnect leaves attribution methods lacking causal validation, while counterfactual methods suffer from severe instability, producing adversarial-like noise instead of meaningful explanations.} In this work, we revisit time-series explainability from an information-theoretic perspective and show that existing explainers are vulnerable to trivial solutions and distributional shifts. To address these limitations, we propose a unified objective function for explainable time series learning that bridges attribution and counterfactual reasoning within a single framework. Building upon the Information Bottleneck principle, our formulation explicitly prevents trivial explanations and out-of-distribution counterfactuals. {Based on this objective function, we introduce {\modelname}, a novel explanation framework that learns a parametric transformation network to construct explanation-embedded instances, where preserved information yields attribution explanations and controlled information removal produces stable counterfactual explanations.} We evaluate {\modelname} on synthetic and real-world benchmarks against state-of-the-art baselines. Extensive quantitative and qualitative results show that {\modelname} consistently outperforms competing methods, yielding faithful attributions and stable counterfactual explanations.

Large language models (LLMs) frequently produce confident yet factually incorrect responses when user inputs contain misleading premises, a phenomenon we attribute to fact perturbations in the input. Existing approaches to hallucination mitigation typically assume reliable user inputs, overlooking how such factual errors can actively mislead model reasoning. To address this vulnerability, we propose DEDUCE, a three-stage framework that transforms LLMs from passive responders into proactive error correctors. DEDUCE operates in three stages: (1) detect errors through fine-grained fact extraction and verification; (2) devise correction strategies via multi perspective deliberation; and (3) correct misconceptions while delivering reliable answers. We also present MisFactQA, a dataset containing factual errors of varying degrees, and propose new metrics for evaluating model robustness. Experiments on TruthfulQA, FalseQA, and our MisFactQA benchmark demonstrate that DEDUCE significantly improves both accuracy and error correction capability. Consistent gains across Qwen, LLaMA, and Gemma families confirm its effectiveness and scalability.

Foundation models have transformed molecular property prediction, yet it remains unclear whether a molecular foundation model, fine-tuned on a single canonical olfactory prediction task, can learn representations that transfer across diverse machine olfaction problems. We investigate this question by fine-tuning Uni-Mol2 on the GS-LF benchmark for multi-label odor descriptor prediction and evaluating the resulting model, without additional deep-learning training, on four complementary downstream settings: cross-dataset odor descriptor prediction, odorous-versus-odorless classification, enantiomer evaluation, and odor mixture discriminability. The fine-tuned model matches or exceeds the performance of the state-of-the-art olfaction-specific baseline on the primary GS-LF benchmark and consistently transfers across these downstream evaluations. The enantiomer analysis further shows that three-dimensional molecular representations distinguish mirror-image molecules in a way that two-dimensional graph models fundamentally cannot, although accurately predicting the perceptual consequences of stereochemistry remains an open challenge. Together, these results support a train-once, transfer-across-tasks paradigm for machine olfaction and suggest that chemically pretrained molecular representations provide a strong foundation for transferable olfactory prediction.

Detailed multicellular growth simulations based on subcellular element models (SEMs) can capture complex tissue development, but their element-level interactions impose substantial computational cost. This work presents a scalable multi-GPU framework for 3D multicellular growth simulation that combines GPU acceleration, spatial binning, domain decomposition, and workload-aware partitioning. Cell movement, growth, and division continuously reshape the spatial workload distribution, causing initially balanced partitions to become inefficient over time. To address this, we introduce an RNN-based load-balancing controller that observes recent per-rank execution times and partition states and learns residual corrections to a reactive boundary-adjustment rule. The controller is trained offline in a differentiable surrogate of the load-balancing loop with randomized workload dynamics, requiring no measured execution traces for training. We evaluate the framework in terms of single-GPU acceleration, multi-GPU computation scaling, controller-level load-balancing behavior, and end-to-end simulation performance, with comparisons against static partitioning, reactive load balancing, and conventional time-series prediction baselines. A representative embryonic epidermal development use case further demonstrates the type of spatially and temporally evolving workload targeted by the framework. In our evaluation, GPU acceleration with spatial binning accelerates the interaction computation by roughly three orders of magnitude over a serial CPU baseline. RNN-guided load balancing reduces the mean global imbalance from 11.3% under static partitioning to 3.5%, lowers end-to-end runtime by 9.0% relative to static partitioning, and reduces slice migration by 7.7x compared with the reactive baseline, showing that history-aware control can improve workload balance while avoiding unnecessary repartitioning.

An intuitive method for dimensionality reduction is proposed, which is highly effective for finding interesting projections of multivariate data. Following similar intuitive motivation to a number of existing techniques, the proposed method is based on enhancing the nearest neighbour relationships in the data. The proposed projection arises from the spectral decomposition of a matrix designed to encode the local covariance structure in the data, where the local covariance at a point is captured by pairs of its nearest neighbours. We show that under standard regularity conditions this matrix is a consistent estimator of the so-called ``Density Information Matrix'' (DIM); a non-parametric analogue of the Fisher Information Matrix. Spectral decompositions of DIMs have been shown to be connected with the important problems of Independent Components Analysis and, in the supervised context, Sufficient Dimension Reduction. However, existing estimators of the DIM are computationally expensive to compute and only target the DIM of a surrogate density, which is proportional to the square of the true underlying density. In addition, we go on to explore the practical utility of our method in aiding the downstream tasks of cluster analysis and outlier detection.

Fine-tuning abstractive summarization models requires high-quality annotated data. However, obtaining such corpora is expensive and time-consuming, as it requires human annotators to read and comprehend long documents to create accurate summaries. Active learning mitigates this issue by selecting only the most informative instances for annotation, allowing models to achieve competitive results with significantly fewer labels. However, the application of active learning to summarization remains under-explored, and existing studies often suffer from instability and significant computational bottlenecks. To overcome these challenges, we propose LOBSTER (LOss-BaSed acTivE leaRning), a novel active learning framework designed specifically for abstractive summarization. LOBSTER improves performance by prioritizing unlabeled instances semantically similar to the model's current high-loss training examples, enabling the model to explicitly correct its specific weaknesses. Our empirical evaluation across three benchmark datasets and two summarization backbone models demonstrates that LOBSTER consistently matches or outperforms current state-of-the-art approaches while achieving a query selection speedup of up to 665x.

LLM pipelines increasingly ingest Office Open XML (OOXML) documents (Word, Excel, and PowerPoint files) as first-class evidence in financial, compliance, and retrieval-augmented workflows, implicitly assuming semantic integrity: that the evidence consumed by the model matches the content shown in the Microsoft Office suite editing canvas. We show that this assumption can fail in OOXML-to-LLM pipelines. The same specification-valid OOXML file can yield one evidentiary view in Microsoft Office and another when extracted for an LLM. Each view is treated as authoritative by its consumer, a condition we call plural ground truth. The ingestion contract rarely states which view and semantic roles become model evidence or preserves how that evidence was derived. We call the specification-grounded OOXML constructions that induce such divergence evidence forks. We systematically traverse and mine the OOXML specification and confirm 21 evidence forks across Excel, Word, and PowerPoint, spanning six dimensions of view construction. All 13 tools in our extraction panel emit evidence from at least one fork. We test four native-ingestion LLM APIs and seven web chatbots. Each test document carries a trap: a task-relevant fact exposed by extraction but not shown in Office. Across this 21-mechanism evaluation, the four APIs return the trap in 48--76% of trials. For 20 of 21 mechanisms, at least one of the eleven interfaces returns the trap. Our measurements further show that exposure is shaped upstream of the model by the ingestion path and extractor configuration. A source-level survey of sixteen popular open-source LLM projects further shows that default OOXML ingestion paths concentrate on affected extractor families.

In federated learning, it is well known that heterogeneous data can (in theory) slow down optimization, and much effort has been directed at designing optimization algorithms that are unaffected by data heterogeneity, such as the SCAFFOLD algorithm. Yet, despite strong theoretical guarantees, SCAFFOLD does not usually outperform the much simpler FedAvg in practice. In this work, we propose that this gap is due to the presence of Edge of Stability (EoS) and progressive sharpening in federated optimization, supported by extensive empirical probing. First, we find that EoS-like dynamics occur with both FedAvg and SCAFFOLD under a variety of architectures and hyperparameters. We observe that the equilibrium value of the sharpness is inversely proportional to the learning rate (as in GD), and interestingly, the degree of data heterogeneity (but not the number of local steps) also affects the equilibrium value. Most importantly, we observe that SCAFFOLD's ability to estimate the gradient of the global objective is severely degraded at the EoS, as measured by the correlation between sharpness and SCAFFOLD's error in estimating the global gradient along the optimization trajectory. This suggests a mechanism for SCAFFOLD's lackluster performance in deep learning: with high sharpness at the EoS, SCAFFOLD cannot reliably estimate the global gradient.

Forecasting in large-scale e-commerce marketplaces is increasingly required to support planning: merchants need to evaluate sales outcomes under future action sequences such as budget schedules, rather than passively predicting what happens next. However, most existing time series forecasting (TSF) approaches remain inherently passive. Even when incorporating operational decisions as auxiliary covariates, they typically optimize for correlation-based extrapolation under historical policies. This design suffers from autoregressive inertia and conflates endogenous market evolution with decision-induced transitions, leading to policy-insensitive rollouts and unreliable counterfactual analysis. To bridge this gap, we propose CEDAR (Controlled and Event-Driven Demand forecasting via Action-aware Residual decomposition), a two-stage framework for robust decision-conditioned simulation. In Stage I, an Action-Interleaved Transformer learns controllable action-conditioned state transitions for rollout under planned interventions. In Stage II, a Residual Correction Module leverages external event signals and LLM-assisted text representations to align noisy event descriptions with product context and correct event-driven deviations. Our study is enabled by a large-scale real-world dataset from Alibaba 1688, comprising approximately 32 million product trajectories with paired state-action sequences and aligned event signals. Extensive offline experiments and online controlled experiments in production demonstrate that CEDAR consistently improves simulation accuracy over strong TSF baselines and delivers practical gains for real-world budget planning.

Large language models (LLMs) increasingly assess generated content, giving rise to the LLM-as-a-Judge paradigm. These systems now score outputs, filter content, and gate iterative refinement in production pipelines, where each judgment is often assumed to be independent of earlier evaluations. We test this assumption using three prompt conditions: no metadata, revision framing, and anchored metadata containing revision, attempt, and prior-score fields. We show that prior scores, even when included only as context metadata, anchor judgments and systematically shift ratings toward their values. Across 192,000 attempted evaluations (185,271 successful), seven out of the eight evaluated models have 95% task-stratified bootstrap intervals below zero for the total anchored-metadata effect on 20 fixed texts. Cohen's $d$, a standardized measure of the difference between score distributions, reaches an absolute value of 0.71. Token-level analysis of selected model-task probes suggests a threshold-like response pattern: introducing anchored metadata produces a marked redistribution of output-score probabilities, while changing the anchor value within the tested below-threshold range produces comparatively little additional variation. On categorical industry data with human-labeled ground truth, anchored metadata blocks 48% of error corrections and flips 10.18% of correct judgments toward an assigned wrong label, demonstrating the bias extends beyond numerical scoring to categorical decisions. Neither Chain-of-Thought nor a metadata-disregard warning reduces the total effect, although the warning improves the paired accuracy effect relative to baseline in the industry experiment. Reliable LLM evaluation demands careful context engineering rather than an assumption of impartiality. Effective mitigation must be validated for the intended model and task or domain.

High-resolution precipitation fields are essential for hydrological impact assessment, yet global climate model outputs are too coarse and biased for direct use. AI-based statistical downscaling with diffusion models offers a promising approach, but the mechanism by which large-scale atmospheric predictors condition generation remains largely unexplored. We investigate three conditioning strategies for a denoising diffusion probabilistic model applied to daily precipitation downscaling: channel concatenation of upsampled coarse predictors, cross-attention conditioning with a learned convolutional encoder, and cross-attention conditioning with the frozen encoder of the pretrained Prithvi WxC weather foundation model. All strategies are evaluated against an unconditioned baseline under identical conditions using probabilistic, distributional, spectral, and extreme-event metrics for the Colorado River Basin. Concatenation conditioning achieves the lowest point-wise CRPS and MSE, but tends to produce over-smoothed fields that suppress high-intensity events. In contrast, cross-attention conditioning provides substantially better distributional realism and modest improvements in spectral fidelity. Improvements are greatest for extremes: the Prithvi-WxC conditioned model retains over half of >100mm/day events, although estimates are uncertain due to limited samples. When trained on the full dataset, the learned convolutional model performs similarly to the foundation model-conditioned approach while requiring lower computational resources. However, the Prithvi-WxC-conditioned model achieves comparable performance with only five years of training data. These results indicate that cross-attention conditioning offers advantages over simple concatenation for probabilistic precipitation downscaling, and that pre-trained foundation model representations may offer benefits in data-limited settings.

Multimodal protein language models (pLMs) learn joint protein sequence-structure distributions, and their generation performance should also depend critically on inference-time sampling strategies. Yet prior work has focused more on model training than on how inference-time strategies behave. In this paper, we establish a three-stage investigation framework to empirically study the inference design space of multimodal pLMs across three representative pLMs and four fundamental tasks. We evaluate vanilla sampling, task-specific classifier-free guidance, and reward-guided beam search on multimodal pLMs, corresponding to controls over sampling distributions, per-step logits, and parallel trajectories. Throughout the complementary advancements centered on exploration-exploitation trade-off, we (1) reveal the suboptimality of default inference protocols and identify task-oriented sampling preferences; (2) observe substantial quantitative gains across tasks, consistently boosting the upper bound performance of multimodal pLMs without updating model parameters; (3) derive conclusions about base models that differ from prior consensus.

Key Point Analysis (KPA) aims to identify a concise set of key points that summarize a collection of arguments together with their prevalence. We argue that KPA is fundamentally a structured prediction problem that requires recovering semantic groupings, generating representative key points, ensuring coverage, and estimating prevalence. Under this formulation, we show that existing KPA benchmarks suffer from limitations in grouping quality, redundancy, coverage, and argument-key point mappings, causing ceiling violation and selection failure in reference-based evaluation. To support future research on true KPA, we introduce a structure-aware, distribution-sensitive benchmark built via a human-in-the-loop re-annotation. Human and LLM evaluations consistently show that the resulting structures yield more coherent groupings, higher-quality key points, better coverage, and more reliable prevalence estimates than existing annotations. We further release several annotation resources to support research on KPA evaluation, argument-key point matching, explainable KPA, and LLM-as-a-judge methodologies, and outline a research agenda for true KPA.

Cough acoustics are promising for non-invasive tuberculosis (TB) screening, yet whether machine learning (ML) models capture disease-related acoustics or artifacts of data collection remains unresolved. We evaluated the cross-dataset generalizability of classical ML and deep learning (DL) cough-based TB classifiers across three independent datasets. Despite moderate within-dataset performance (ROC-AUC up to $0.755 \pm 0.056$), both pipelines fail to generalize, with external performance frequently below 0.6, indicating a possible limitation of the data. We further observed audio representations are organized by recording device and dataset rather than TB status, predicted TB probability tracks country-level prevalence in CODA, and device mismatch degrades transfer while device-diverse training improves it. Additionally, a clinical-variable baseline generalizes more consistently (ROC-AUC $0.655 - 0.711$), indicating acquisition-specific variability is a stronger driver of poor generalizability than population shift. High within-dataset performance is not enough. External validation is essential before cough-based TB models are clinically ready.

Drug synergy prediction estimates whether two drugs produce a stronger joint effect than expected from their individual activities. For drug combination discovery, a single synergy score is often not enough: researchers also need to know which molecular regions jointly drive the prediction. We study motif-pair synergy explanation, which identifies pairs of chemically coherent regions, one from each drug, that jointly contribute to predicted synergy. Existing interpretable synergy models expose atom- or substructure-level signals, but their explanations are built into the predictor architecture, and none validates cross-drug region scores under repeated perturbations or feeds that evidence back to refine the explanation. A reliable motif-pair explanation should instead be chemically coherent, perturbation-stable, and aligned with predictor behavior. We introduce VINCENT (Validated Interaction Network for Cross-drug Explanation of Therapeutics), a post-training framework for a fixed interaction-aware synergy predictor. VINCENT extracts atom-pair evidence from attention and gradient signals, groups atoms into chemically coherent motifs, and validates candidate motif pairs through repeated local perturbations. The validated evidence is fed back to refine motif assignments, yielding explanations that satisfy these three criteria. On a 25-pair literature-annotated subset, VINCENT achieves a mean motif recall of 0.826 (95% CI: 0.78-0.87), compared with 0.49-0.66 for baselines. Across all 71 test pairs, its validated interaction scores yield a TP/TN separation of 3.36. These results show that closed-loop perturbation validation recovers literature-supported molecular regions more accurately than existing alternatives while producing cross-drug interaction scores that better reflect predictor behavior.

AI weather prediction (AIWP) models rival physics-based models, yet the sources of their unexpected forecast accuracy and the degree of their physical fidelity remain unclear. Here, across a hierarchy spanning observation-based reanalysis, a general circulation model, and the multi-scale Lorenz system, we show that AI models can be trained to skillfully predict the past (backcast), though backcasts are systematically less accurate than forecasts. However, skillful backcasting appears to violate the second law of thermodynamics, and all these forecasting and backcasting models miss the butterfly effect. We trace the surprising forecast accuracy, missing butterfly, and skillful backcasting to a single cause: inevitable coarse-graining of training data, which removes fast, small scales and/or some variables. From the Lorenz system to official Pangu-Weather models, reducing coarse-graining makes AI predictions more physics-like (arrow of time and butterfly-like effects emerge), but forecast accuracy declines. Results offer an explanation for AIWP models' forecast skill: unlike physics-based models, they implicitly learn how fast, small scales affect large scales without inheriting their rapid error growth. Broader implications are that AI models' proliferation calls for revisiting predictability theories and long-term climate emulation strategies, and backcasting offers a useful, new lens for such analyses.

Large language models access knowledge inconsistently across languages, but to what extent do they differ in their skill sets when interacting with different languages? This work quantifies cross-lingual skill inconsistency orthogonally from knowledge and general benchmark performance. We do this via multilingual self-play: two instances of the same model compete in a text-based game, each interacting through a different language interface. Since the model, opponent, rules, state space, and available actions remain fixed, this setting isolates the effect of language on the model's realized behavior. We build a multilingual extension to TextArena and evaluate three open-weight models across eight languages and six games covering spatial reasoning, imperfect information, resource allocation, and repeated interaction. We find that the same model can exhibit markedly different playing strength across languages, with systematic variation in win--loss margins, invalid actions, and strategic tendencies. Detailed analyses reveal language-specific failures in spatial reasoning, card-conditioned decisions, and optimal move selection. In some settings, changing only the intermediate reasoning language recovers much of the lost performance, suggesting that language can affect different stages of the decision process. These results show that skill discrepancies are a measurable major roadblock in the development of truly multilingual models. Better understanding these discrepancies can help us design models that perform more equitably across languages.

We present FlowMoDL, an unrolled neural network for highly accelerated 4D flow MRI reconstruction that directly optimizes for both anatomical magnitude and phase-derived velocity accuracy. Building on the MoDL framework, FlowMoDL alternates a learned (3+1)D spatiotemporal denoiser with conjugate-gradient data-consistency updates based on the SENSE forward model. A novel dual-pathway conditioning scheme adapts the denoiser features and data-consistency weighting, enabling a single model to handle varying acceleration factors ($10\times$ to $50\times$). To ensure physiological accuracy, the network is trained using a deep-supervision composite loss that explicitly penalizes velocity magnitude and angular errors, stabilized by a curriculum schedule. We evaluate FlowMoDL on the multi-center CMRx4DFlow dataset against classical and deep-learning baselines (CG-SENSE, MoDL, FlowVN, and FlowMRI-Net). A key advantage of FlowMoDL is its superior gradient step efficiency. When evaluated under an equivalent, limited budget of gradient steps, competing flow-specific networks degrade significantly. In contrast, FlowMoDL robustly converges and strictly outperforms all competitors across all acceleration factors in magnitude SSIM, nRMSE, relative velocity error, and angular error, successfully recovering sharp structural details and temporally coherent velocity fields.

A recent line of synthetic-data work reconstructs the thinking behind existing text rather than rewriting the text itself, but it operates on short web passages, recovers only local thoughts, and leaves the structure of whole documents untouched. Scientific papers are written to a clear and largely uniform structure and make a natural substrate for lifting this paradigm to the document level. We present a pipeline that unfolds each paper into a multi-turn generation trajectory in which a teacher model reconstructs the writing process of the whole paper: a writing request, a global plan, and pre-writing deliberation for each section. All section texts and the abstract are kept verbatim from the source paper. We apply the pipeline to quality-filtered arXiv papers and obtain a corpus for continued pre-training (CPT) that is roughly twice the size of the source text. The same reverse construction extends to instruction data and evaluation. Treating real paper text as the answer yields an SFT dataset. Anchoring tasks in held-out papers yields PAW-Bench, an academic-writing benchmark whose tasks carry their own rubrics and checklists. In controlled experiments CPT on our corpus followed by supervised fine-tuning on public datasets improves writing benchmarks broadly while preserving general reasoning and improving long-document reading. The writing gain persists even when every model is fine-tuned on a dedicated writing SFT dataset. Mixing our SFT data into that recipe lifts academic writing further.

Large language models (LLMs) are increasingly used as evaluators to assess output quality and preference alignment, yet providing reliable guarantees of agreement with human judgments remains challenging. Recent work introduces confidence-thresholding methods that provide such guarantees for pairwise comparisons, relying on the assumption that higher estimated confidence implies lower disagreement risk with humans. However, this assumption can break down when the number of candidate responses increases, since distributing probability mass across many alternatives can distort confidence estimates. To address this issue, we propose a Localize-Then-Decide framework. First, conformal prediction localizes a small shortlist that contains the human-preferred response with high probability. Then, a calibrated confidence-based rule selectively chooses a single response from this shortlist or abstains. This design restores the monotonic relationship between confidence and disagreement risk and enables high-probability agreement guarantees. Experiments with multiple candidate sizes across several datasets and judge LLMs demonstrate that our framework consistently achieves higher guarantee success rates and substantially higher coverage than single-stage baselines.

Accurate regional near-surface temperature forecasting is fundamental to short-range weather services and downstream risk assessment. Existing deep learning-based regional forecasters commonly produce a fixed set of future frames on a prescribed grid, limiting their use when forecast products must be evaluated at query-dependent lead times or display resolutions. To overcome these fixed-output constraints, we formulate regional T2M forecasting as query-conditioned continuous spatiotemporal temperature field evaluation and propose the Continuous Spatiotemporal Temperature Forecaster (CSTF), a neural field that turns forecast lead time and output resolution into explicit queries when evaluating 2-m temperature (T2M). Specifically, CSTF first encodes multivariable ERA5 histories into latent meteorological states and then decodes T2M as a coordinate-based field. Accordingly, spatial location, forecast lead time, and output resolution are introduced as queries, enabling standard hourly forecasts, intermediate lead-time diagnostics, and resolution-controllable outputs within a unified field-evaluation framework. Furthermore, to maintain coherence across flexible field queries, we design spatial-gradient, temporal-difference, and scale-consistency objectives that regularize regional thermal structures, lead-wise evolution, and cross-resolution agreement. Experiments on the Southeast China 0-6 h ERA5-Land benchmark demonstrate that CSTF achieves the best aggregate deterministic skill, including a 17.0 percent reduction in Bias, with global-scope diagnostics further illustrating flexible lead-time and resolution-controllable inference.

For overparameterized neural networks, many solutions can fit the training data equally well while behaving very differently on unseen samples. Grokking separates training fit from visible generalization, providing a window for studying how this selection develops during training. We sweep short, fixed-duration weight-decay (WD) pulses across this plateau and measure how they shift later generalization time. Across three grokking tasks, these shifts are unordered early in the plateau but later form a stable dose ordering, with stronger WD increases leading to earlier generalization and stronger WD decreases leading to later generalization. This ordering emerges before visible generalization in all three tasks. Meanwhile, test-loss barriers between perturbed and baseline generalization checkpoints collapse toward zero while the ordered timing effects persist. We call this combination of increasingly constrained solution selection and persistent dose-ordered timing sensitivity the canalization of function selection.

Kolmogorov-Arnold Networks (KANs) replace fixed activations in deep architectures with learnable univariate edge functions, making the choice of edge parametrisation central. Existing variants rely on fixed bases such as splines, polynomials, or Fourier features, which impose a function-space geometry before data are observed. We introduce geometry-constrained KANs, a family of edge activations derived from Banach duality maps in which the geometry itself is learned through a scalar exponent $p > 1$ per edge. This exponent controls the qualitative response: sub-Euclidean values produce sharp, threshold-like behaviour reminiscent of the $\ell_1$ (LASSO) geometry, $p = 2$ recovers the linear regime, and larger values produce flatter responses near the origin. Across 50 symbolic-regression targets ($40$ from the AI Feynman benchmark plus $10$ synthetic stress tests), geometry-constrained KANs match or beat every fixed-basis baseline on median NRMSE (Banach-KAN $0.030$, tying Chebyshev and improving on splines); on average rank Banach-KAN is best on the $18$-equation core ($2.00$) and statistically tied with the strongest spline on the full benchmark ($2.32$ vs. $2.34$). The clearest gains appear under measurement noise: as $σ$ grows from $0$ to $1$, $\ell^p$-KAN degrades only $3.7\times$ -- below even a cross-validated spline ($\approx 11\times$) -- while an unregularised spline degrades $21.6\times$; Banach-KAN degrades $8.8\times$, comparable to a tuned spline but far more stable than the unregularised one. Banach-KAN also takes the most per-equation wins in the small-sample regime, with fixed-basis models catching up only as the training set grows. Learned exponents provide an interpretable, relative signal: at a fixed initialisation they reveal a consistent, target-dependent geometric ordering across equation families and input dimensions.

Contact-rich manipulation requires adapting to contact states that can evolve substantially within an action horizon. However, chunk-based vision-language-action models predict complete action chunks from observations collected before execution, leaving tactile conditioning stale during execution. Existing tactile-reactive approaches typically rely on separate high-frequency controllers, which increase both architectural and training complexity. In this paper, we introduce TacForcing, a streaming action-generation framework that effectively incorporates execution-time tactile feedback. Instead of employing a separate reactive controller, TacForcing replaces the standard action expert with a streaming action expert to generate actions conditioned on the evolving tactile observations acquired during execution. TacForcing also introduces Execution-Aware Tactile Attention (EATA), which restricts tactile conditioning to actions nearing execution, thereby reducing the temporal mismatch between tactile acquisition and action execution. Across six simulated UniVTAC tasks and three real-world contact-rich manipulation tasks, TacForcing achieves average success rates of 65% and 69%, respectively, outperforming strong baselines in both settings.

Federated Learning (FL) enables distributed training of machine learning models while preserving data privacy. However, FL struggles with heterogeneous, non-IID client data distributions, resulting in sub-optimal and biased global models. In this paper, we propose pFedMARL, a novel approach leveraging Multi-Agent Reinforcement Learning (MARL) with Twin Delayed Deep Deterministic Policy Gradient (TD3) to dynamically adapt aggregation strategies in FL settings. Our method employs a server-side agent adjusting client contributions to optimize global model robustness and client-side agents balancing global and local updates to personalize models effectively without pre-training. We demonstrate superior performance of pFedMARL for training a semi-supervised audio spectrogram transformer, matching or outperforming FedAvg, Ditto, and local training approaches across multiple non-IID scenarios and in the presence of adversarial clients. Our results indicate that pFedMARL actively improves accuracy, robustness, and fairness, making it suitable for real-world deployments.

Containerization dominates enterprise software delivery, yet Dockerfiles that assemble container images frequently harbor security misconfigurations and structural technical debt. This problem is poorly understood in corporate inner-source environments, where proprietary context and isolated governance prevent direct application of open-source findings. We present an automated, six-stage pipeline that: (1) crawls an enterprise GitLab instance, (2) enriches each Dockerfile with static security and quality metrics (Hadolint, ShellCheck, Trivy) and lifecycle data, (3) groups functionally identical workloads using LLM-generated semantic descriptions and HDBSCAN, and (4) quantifies the optimization gap against cluster-internal reference implementations. Applied to 11,470 Dockerfiles from over 6,200 repositories at a single large industrial company, we find a systemic deficit: 99\% of files contain at least one security misconfiguration, 80.8\% violate Dockerfile best practices, and the median artifact has not been revised for 838~days. Despite this, high-quality reference implementations already exist within 83\% of functional clusters. Adopting these internal standards would increase the average security posture score by 60.4\% without developing any new templates. These findings, grounded in one organization's inner-source ecosystem, provide a data-driven foundation for future automated, context-aware recommender systems targeting enterprise supply-chain security; whether the observed technical-debt distribution and optimization gap generalize to other enterprises remains an open question for future multi-organization study.

Modern GUI-agent frameworks achieve strong desktop task performance with frontier API models, yet persistent control information often remains implicit in growing interaction trajectories. At each step, the planner reconstructs the active task stage, accumulated evidence, and runtime feedback before deciding the next action. This dependence becomes more pronounced under weaker local reasoning backbones. Across four representative state-of-the-art frameworks, replacing GPT-5 with Qwen3.5-9B reduces average OSWorld SR-100 from 60.9\% to 37.7\%. Trajectory annotation further identifies at least one control failure in 91.6\% of failed trajectories. To address this problem, we introduce LocalLSTC, a training-free architecture that organizes control by temporal scope, maintaining persistent cross-step state to guide short-term execution commitments. Long-Term Control maintains the active subgoal, subgoal-aligned evidence, and runtime feedback across interactions, while Short-Term Execution realizes bounded commitments for the current step. Long-to-Short Planning forms each commitment from persistent state, and Short-to-Long Control integrates execution outcomes back into that state for progress assessment, recovery, and termination. With Qwen3.6-27B, LocalLSTC reaches 64.7\% SR-100 on OSWorld and 65.3\% on WindowsAgentArena, outperforming the strongest prior local results on both benchmarks. Ablations further support contributions from mechanisms on both sides of execution. These findings identify temporal organization of control information as a distinct architectural dimension for locally deployed GUI agents.

Self-evolving LLM coding agents write their own tools by imitating retrieved skills from shared skill libraries. We identify a vulnerability in this loop: during authoring, a retrieved malicious skill can become the template for a new skill that preserves the payload. We call this self-poisoning: the agent authors, stores, and runs the resulting malicious skill. We exploit it through EvoMal, an attack that amplifies self-poisoning by wrapping an interchangeable payload in a banner, a set of benign-looking structural elements that induces an imitating agent to reproduce the enclosed code. The attacker plants malicious skills in the library without invoking them. The agent then authors and executes new skills carrying the harmful code. Each authored copy can re-enter the library and be imitated again, forming a self-propagating worm that persists after the planted skills are removed. We define the agent self-poisoning rate (ASPR) as the fraction of tasks that add a newly authored malicious skill to the library. Across six models on 153 tool-relevant SWE-bench Verified tasks, ASPR ranges from 20.3% to 41.8%, and the poisoned libraries hold 4.9 to 9.0 times as many malicious skills as were planted. The vulnerability also appears without a banner: DeepSeek-V4-Pro reaches 11.1% ASPR with the payload alone. Tailoring the planted skill descriptions to one task family raises ASPR to 86.7%. After the planted skills are removed, Qwen3 retains a round-5 ASPR of 68% because agent-authored copies remain. These copies evade existing defenses, which focus on attacker-submitted names, code, and signatures. We propose counter-prompt, a defense that discourages banner-style copying and reduces EvoMal's ASPR to at most 6.7% with no significant task-completion loss.

Large Language Models (LLMs) exhibit strong problem-solving abilities, positioning them as promising agents for Socratic teaching to guide students through step-by-step heuristic questioning. However, existing approaches typically adopt a one-problem-one-solution paradigm, restricting the teaching guidance to a single linear reasoning path. This design limits instructional flexibility, weakens error recovery, and restricts students' ability to engage in parallel thinking to explore multiple valid solutions. To overcome these, we propose ToST, a Tree-of-Thought Socratic Teaching framework that explicitly supports multi-path guidance under a one-problem-multiple-solutions paradigm. ToST employs Parallel Sowing, a parallel-thinking-oriented questioning strategy to encourage students to approach problems from diverse perspectives, and a Multi-Path Adaptive Guidance mechanism to provide more robust and non-linear instructions across alternative solution trajectories. Concurrently, to fill the void in systematically evaluating such non-linear instructional capabilities, we advance the task of multi-path Socratic guidance by establishing MPSG-Bench, a comprehensive benchmark that includes a dataset of 31K multi-path teaching dialogues and a five-dimensional evaluation framework grounded in the SOLO (Structure of Observed Learning Outcomes) theory to assess parallel-thinking guidance. Experimental results demonstrate that ToST significantly enhances guidance success rates while empowering students to navigate and explore multiple solution paths more effectively under both automatic and human metrics.

EXAONE Tabular is a compact tabular foundation model family for classification and regression via in-context learning, producing predictions without dataset-specific gradient updates. Pretrained exclusively on a synthetic structural-causal-model (SCM) prior, its central contribution is an architecture-centered redesign of tabular in-context learning. Rather than compressing features into a fixed row embedding before a separate row-level learner, EXAONE Tabular interleaves feature-axis attention within each item with support-conditioned item-axis attention within each feature at every Transformer layer, mediated by item-summary and feature-summary tokens. Across four public benchmarks, EXAONE Tabular combines strong predictive performance with high efficiency. On TabArena, its 20.81M-parameter classification model ranks first overall, surpassing tuned ensembles and 4-hour AutoML pipelines, while regression reaches the performance regime of the 1.64B-parameter TabFM at roughly 1/11 the inference cost. On BCCO and TALENT, EXAONE Tabular ranks second in classification and first in regression. On ScoringBench, it achieves the best mean rank for both point-estimation and predictive-distribution quality, leading the $R^2$, RMSE, and CRPS evaluations. Together, these results establish EXAONE Tabular as a state-of-the-art compact tabular foundation model family, combining strong predictive performance across classification, point regression, and probabilistic regression with an efficient model design.

Understanding user preferences from noisy and temporally evolving social media behaviors is fundamentally challenging due to interest drift, where user preferences shift across time and exhibit both multi-scale temporal patterns and diverse co-existing interests. To address this, we propose DUMoE, a unified framework for drift-aware multimodal user representation learning. Our model consists of (i) a temporal dynamics-aware backbone that captures and integrates static profiles, short-term behavioral signals, and long-term dependencies into a coherent representation, and (ii) a sparse mixture-of-experts (MoE) interest adapter that disentangles multiple latent interests via expert specialization and adaptive routing. Each expert models a distinct interest subspace, while a gating network dynamically selects and aggregates a sparse subset of relevant experts for each user. To enable stable and effective optimization, we further introduce a three-stage training strategy that decouples backbone learning, expert specialization, and gating optimization. Extensive experiments on real-world social media datasets show that DUMoE consistently outperforms state-of-the-art methods on both user interest prediction and interaction prediction tasks.

In serious illness, human surrogates often struggle to accurately predict patient preferences (68% accuracy), causing decision conflict. Personalized Patient Preference Predictor (P4) agents offer a potential solution, but prior prototypes treat values as static ratings, ignoring the contextual, situation-dependent nature of medical choices. Grounded in the 'logic of care', we present P4-DT (Dilemma Training), a P4 agent that constructs a patient decision policy by engaging users with varied medical dilemmas, eliciting individual preference reasoning through bi-directional training. In a study with 12 patient-surrogate dyads, P4-DT predicted patient treatment choices with 81.7% accuracy, significantly exceeding chance (OR = 5.61 [2.03, 15.51], p < .001) and outperforming both unassisted surrogates (55.0%; OR = 3.67 [1.59, 8.47], p = .002) and surrogates assisted by P4-DT (61.7%). Comparative prompt analyses showed that incorporating contextual scenario decisions and open-ended text improved accuracy by 15.0 percentage points over initial values ratings alone. We discuss implications for further testing and designing of context-aware AI agents that embody richer human experience to partner in complex decision-making.

This paper investigates the feasibility of early detection of struggling students during CS1 programming exercises using keystroke-level logs. Some students fail to reach a correct solution before the exercise ends, and by the time this becomes apparent from grades or final outcomes, the opportunity for timely instructor support aimed at helping them recover may have passed. We use data from the CodeBench platform, which records real-time code editing events at the keystroke level, alongside execution and submission logs. We define two outcome groups: Breakthrough (BT) students, whose prior submissions all receive 0% and whose final submission achieves full credit, and Fully Stuck (FS) students, whose submissions all receive 0% without reaching a correct solution. To examine this feasibility, we focus on two questions: (RQ1) whether adding keystroke-level editing features improves the prediction of FS students over execution-log features alone, and (RQ2) at which stage BT and FS students can be predicted most accurately. Experiments on the 2019-1 semester of the CodeBench dataset, comprising 507 students, compare three feature configurations: execution-based features (ExecOnly), CodeMirror-based features (CMOnly), and their combination (Combined). We evaluate prediction across successive submission-based stages during each exercise. In the earliest stage, CMOnly outperforms ExecOnly (AUROC 0.654 vs. 0.575), and Combined further improves over ExecOnly by +0.098 (AUROC 0.674). Across all configurations, the earliest stage yielded the strongest predictive signal. These findings indicate that behavioral signals present at the very start of an exercise contain useful clues about whether a student will ultimately solve the problem, and that keystroke-level editing logs provide additional value for early prioritization beyond execution logs alone.

Turkish encoder models have adopted modern architectures while leaving the pretraining objective fixed at masked language modelling. This paper introduces MoganBert-TR, a 149M-parameter Turkish encoder foundation model trained from scratch on a language-specifically filtered corpus, together with an embedding model derived from it (MoganBert-Embed). MoganBert-TR is trained over 237.3B tokens with a two-stage CLM-to-MLM curriculum: causal language modelling first, masked language modelling for the remainder, with the transition made inside the stable phase of a WSD schedule. In a controlled ablation under an equal step budget, this design outperforms pure MLM by 2.7-3.7x on Turkish MS MARCO retrieval; the measured mechanism is embedding geometry, where a single direction absorbs 28.1% of the variance under pure MLM against 11.9% under the curriculum. Long-context extension and learning-rate decay are then split into two branches after a shared prefix: running the final portion of decay at 1024 context improves the TrGLUE average by 0.49 +/- 0.26 points across five paired seeds (p = 0.013) and beats a model-soup alternative by 0.75 points at ~4.3% additional cost. MoganBert-TR attains 78.41 on TrGLUE, the best among the Turkish ModernBERT models compared, and 77.73 on TabiBench, where it leads two of the eight categories with the largest margin on code retrieval (+3.62 points over TabiBERT). MoganBert-Embed, produced through teacher distillation and multi-signal contrastive fine-tuning, ranks first among student models on the MTEB(Turkish) overall average with 68.30 and reaches 99.5% of its 7.57B-parameter teacher's score with a 51x smaller backbone. The accompanying 50,048-token tokenizer outperforms all compared Turkish tokenizers on compression and fertility across two independent test sets. Weights, tokenizer, embedding model and evaluation code: https://huggingface.co/moganai

One common trade-off in the use of large language models involves reducing the size of the model while increasing the amount of computation at inference time, for example by using a wider beam search. In this paper, we examine the constrained case of this "model size vs. inference compute" trade-off, in which the model outputs are constrained by a strict grammar at inference time. Our results demonstrate that the constrained trade-off behaves differently from the unconstrained trade-off. We investigate the task of converting a prose query into an equivalent SQL query (text-to-SQL). Performance is evaluated on the Spider text-to-SQL benchmark, using the Qwen2.5-Instruct model family ranging in size from 0.5B to 7B parameters, all at 4-bit precision. We experiment with two approaches to varying inference compute: (i) beam search with a variable number of beams; and (ii) sample+vote, i.e., sampling several constrained outputs and then voting on their execution results, where the number of samples is varied. On the 1034-example development set, we find that: (a) both beam search and sample+vote improve accuracy, especially on smaller model sizes; (b) the "model size vs.\ inference compute" trade-off is not advantageous in this experiment, because moving to a larger model size typically results in higher accuracy than increasing inference compute on the same model size; (c) beam search outperforms sample+vote at a matched inference budget. This latter result is of particular interest since it contrasts with the findings of the unconstrained trade-off.

The inappropriate disposal of solid waste remains a significant public health and environmental concern worldwide, including in Ghana. Poor sanitation and improper waste management practices contribute to substantial economic costs and avoidable deaths annually. In 2022, a field study in Atonsu, Kumasi, Ghana, reported a community-perceived relationship between household waste disposal and illness patterns, but only through descriptive analysis without quantitative validation. This study extends that investigation using two data-driven approaches. First, a Random Forest classifier was developed to predict illness categories using waste disposal practices and demographic survey data. On a held-out group of respondents who reported illness (N=69), the model obtained a macro F1 score of 0.63, with disposal method emerging as the most important substantive predictor of illness type. Second, a MobileNetV2 image classification model enabled automated waste sorting via visual recognition, achieving 88.2% accuracy and a macro F1 score of 0.87 on the test set (N=415). The vision-based approach offers an affordable, camera-driven alternative to complex multi-sensor systems, making it highly suitable for resource-constrained settings. Taken together, the findings provide quantitative evidence for a community health relationship previously documented only qualitatively. They demonstrate the potential for automated waste-sorting in low-resource environments. Importantly, the results illustrate that technological performance alone does not guarantee public health improvements; effective institutional support and implementation are equally necessary.

Generalist vision--language--action (VLA) policies learn long-horizon behavior mainly through short-horizon action prediction and reveal little beyond sampled commands. This creates two coupled bottlenecks: a single action target must implicitly absorb task progress, intermediate intent, and local reliability, while these control states remain hidden during execution. Inspired by functional principles of biological sensorimotor control, we introduce LM-X , which organizes prediction across task, event, and motor scales without claiming anatomical correspondence. Three explicitly supervised signals are emitted online and directly condition action generation: return-to-go (RTG) measures visible task progress, event-to-go (ETG) identifies the next semantic transition, and heteroscedastic action flow estimates local reliability through propagated variance. Explanation is therefore intrinsic to control rather than generated post hoc. Before a costly 20-day pretraining run on 64 NVIDIA B200 GPUs, a controlled five-task pretraining gate verifies the design: the complete model improves success by 16.0 points over the action-only backbone and by 10.8 points over the strongest single-head variant. We then train LM-X on more than 20,000 hours of real-robot trajectories, including over 1,000 hours of failed policy rollouts. LM-X achieves 74.1\% across 50 randomized-hard RoboTwin2.0 tasks versus 55.4\% for GR00T N1.7, and 68.6\% versus 50.7\% across seven real-robot tasks. RTG tracks semantic progress and visible regression, while variance rises during hesitation and oscillatory control. These results show that explicit multi-timescale predictive state can strengthen control while exposing interpretable internal estimates.

Reinforcement learning post-training drives reasoning and agentic capabilities in modern AI systems, yet a growing body of work shows that it is most effective when used to fine-tune an already capable base model. We question whether existing pipelines yield models that are most suitable for reinforcement learning. Building on prior work highlighting the role of coverage and pass@K as predictors of post-RL performance, we design a simple modification to supervised fine-tuning, TailSFT, which filters out already fit sequences during training, thereby focusing learning on under-modeled regions, or the tail, of the data distribution. We justify and validate the design choices in TailSFT, particularly the specific filtering criteria, through a combination of controlled experiments and theoretical analysis. On OLMo-3 7B, TailSFT often improves pass@16 performance on math and coding evaluations, with gains up to 17% absolute, while incurring minimal computational overhead. These higher-coverage checkpoints consistently translate to up to 4% absolute pass@1 gains in subsequent GRPO runs, demonstrating that TailSFT checkpoints serve as better initializations for RL. We further introduce a lightweight diagnostic for identifying settings where TailSFT is most likely to help. More broadly, our results motivate a principled, stage-aware approach to model development, in which intermediate checkpoints are judged by how effectively they support subsequent training.

A key result in statistics is the data processing inequality, originally proved by Blackwell (1951) and later refined by DeGroot (1962) in terms of statistical uncertainty. It states that the Bayes risk of a statistical experiment obtained by stochastically modifying another experiment cannot be lower than the Bayes risk of the original experiment, regardless of the loss function or prior chosen. In machine learning, this result underlies applications such as the information bottleneck principle and some feature learning techniques. However, machine learning problems are constrained learning problems: the model class used does not include all measurable functions. We present a simple counterexample showing that the classical data processing inequality fails to hold in such a setting. Hence, we formulate a generalized data processing inequality, requiring the constrained Bayes risk of a joint distribution (with respect to a loss function and a constrained hypothesis class) to lower bound the constrained Bayes risk on the stochastically modified distribution, regardless of the choice of distribution. We show this inequality to be equivalent to a set containment condition on a specific function set induced by the loss and model class, called the superprediction set. Finally, we derive sufficient conditions for this containment.

Neural operators applied to transient-dynamics PDEs with strong discontinuities exhibit autoregressive instability: in concrete-penetration stress-field prediction, the wavelet neural operator (WNO) diverges in autoregressive rollout, while MeshGraphNets collapse to zero predictions. WNO's instability stems from the lack of a structural constraint on the spectral radius of its propagation operator; the Fourier neural operator (FNO) is stable in these measurements but only emergently, not by construction. We propose a constitutive Markov physics-informed neural operator (MPNO) modeling one-step evolution as a Markov (row-stochastic) propagation operator. Physics-coupled edge weights (acoustic-impedance harmonic mean, contact area, and traction amplitude) encode material-interface constitutive information into a nonnegative symmetric adjacency matrix W; after normalizing the graph Laplacian L = D - W by lambda_max, the propagator P = I - alpha*L~ is constructively constrained to spectral radius rho(P) <= 1, suppressing exponential amplification of autoregressive errors. Stability is thus a designable architectural property, not an optimized loss objective. On three PDEs (Burgers and two-dimensional transverse-section concrete penetration), MPNO rolls out stably with bounded error on all test seeds at 100/135/165 m/s; the single-step relative L2 error is 0.7304 +/- 0.0008, better than WNO and comparable to FNO at about one quarter of FNO's parameters. The edge-weight formula transfers across scenarios by replacing material-property variables. With about 20K parameters, MPNO delivers roughly 10^5x inference speedup over LS-DYNA.

Graph neural networks (GNNs) are widely used to represent complex interactions and relationships among entities. We investigate a multimodal model that combines two complementary ideas: a self-supervised method that enables a GNN encoder pretrained on one dataset to operate directly on another dataset with a different node-feature dimensionality, without rebuilding the model or realigning the data; and an alternating optimization method that updates a language-model module in an E-step and a GNN module in an M-step, rather than jointly training a large language model and a GNN end to end on a large graph. Despite expectations, the combined model did not sufficiently improve predictive performance. We identify six factors: (1) an external anchor in the E-step has a strength-safety trade-off: a weak anchor has little effect, whereas an overly strong anchor can damage the graph representation; (2) the knowledge of the E-step teacher is not injected directly into the GCN embedding Z; (3) the representation space constructed in the M-step is not optimized for the same objective as the E-step teacher space, resulting in a compromise representation for target classification; (4) GCN propagation averages a node's own textual information with information from its neighbors; (5) cosine alignment does not guarantee axes that are discriminative for classification, so stronger geometric alignment with the E-step text anchor need not sufficiently improve the target decision boundary or classification performance; and (6) the force that preserves the source-side self-supervised geometry in the M-step conflicts with the force that moves the representation toward the E-step teacher. We support these observations through a staged set of experiments that varies the influence of the E-step.

Spiking Neural Networks (SNNs) are increasingly distributed as pretrained checkpoints and reused as backbones for new tasks. However, current SNN watermarks are mainly verified against the model output. Thus, a user who replaces the output head can keep most of the original network while removing the evidence used for verification. We present MeMark, a watermark designed for the checkpoint-reuse setting. Instead of storing the watermark in the output head, MeMark embeds a multi-bit identifier in the internal membrane state of selected Leaky Integrate-and-Fire (LIF) neurons. A secret input drives each selected neuron to the chosen side of its own firing threshold, and the same threshold is later used to recover the secret bit, so the verifier does not need a learned decoder. We evaluate MeMark across recurrent, convolutional, residual, and transformer SNNs. On a 215.4M-parameter SpikeGPT checkpoint, all 20 independent 64-bit keys pass the fixed 51/64 verification rule, while none of the $30\,000$ fresh random keys pass when tested against all 20 protected checkpoints and the clean model. All 20 genuine keys also remain above the threshold after fine-tuning, 90\% pruning, int8 quantization, and output-head replacement. Under our stated threat model, adaptive attacks can weaken the watermark but do not remove the ownership evidence in the settings we test. Additionally, we study false ownership claims, key-aware and key-agnostic removal, partial key disclosure, rollback, and extraction into a student. The results show that MeMark can provide evidence of checkpoint derivatives, while being resistant to the adversary's attacks and complete head replacement.

Hosted retrieval-augmented generation (RAG) and semantic search allow users to query valuable provider-held corpora, raising two competing demands: to hide each query and chosen result, yet reveal only the documents that the user is authorized to receive. Existing cryptographic approaches either make this costly by processing the entire corpus for every query, or sacrifice quality for efficiency by scanning a few clusters. We repurpose learned deep hashing as a private filter: a randomized binary code points the provider to a short candidate list, while encrypted reranking and oblivious key transfer protect the precise query and final selection. This shortlist short-circuits full-corpus cryptographic search without sacrificing retrieval quality: with 200-500 candidates, it closely matches full-corpus retrieval across five zero-shot corpora spanning 25K to 5.4M documents. On the full 2.68M-passage NQ corpus over a 10-Gbps link, our protocol only adds 0.73 seconds, or 10 percent, to a 128-token Qwen3-32B RAG pipeline. The released code satisfies directional metric differential privacy (DP) and substantially reduces embedding-inversion and property-inference leakage, demonstrating that a carefully learned shortlist can make private dense retrieval both accurate and practical.

Large language models (LLMs) have recently advanced graph neural networks (GNNs) by enriching node representations with semantic information, giving rise to LLM-enhanced GNNs that achieve substantial performance gains. However, their vulnerability to privacy attacks, in which adversaries infer sensitive information from model outputs, remains largely underexplored. To bridge this gap, we present a systematic evaluation of privacy risks in LLM-enhanced GNNs through a unified framework consisting of five stages: (1) dataset preparation, (2) victim model training, (3) privacy attack, (4) risk assessment, and (5) defense analysis. Specifically, we conduct experiments on six real-world text-attributed graph datasets covering diverse domains. We consider six representative privacy attack methods targeting three fundamental threats, namely link, label, and membership inference, and construct 42 victim model configurations by combining multiple LLM-based feature enhancers with representative GNN backbones. Extensive experiments show that, despite their utility improvements, LLM-enhanced GNNs consistently exhibit increased vulnerability to privacy attacks compared to shallow text representation baselines. Further analysis reveals that semantic enrichment amplifies link-, label-, and membership-related signals in the embedding space, making them more exploitable by inference attacks. Finally, we evaluate differential privacy as a defense strategy and show that, while it can partially mitigate privacy risks, it introduces significant utility degradation, highlighting a fundamental privacy-utility trade-off in LLM-enhanced graph learning. Overall, this work provides a comprehensive understanding of privacy risks in LLM-enhanced GNNs and offers practical insights for developing more secure and trustworthy graph learning systems.

Time is of the essence when dealing with multiple reward signals and non-linear utility. In this paper we argue that the current main approaches in multi-objectiveRL (SER and ESR), and successor features, are insufficient. While each approach deals with non-linear effects on user utility on different timescales, none of them take into account that different effects happening on different timescales can happen within the same decision problem. We motivate that this can indeed be the case by an example, both intuitively and numerically, leading to a new perspective, and a significant and non-trivial gap in the literature.

Retrieval-augmented generation (RAG) is widely assumed to mitigate factual errors in large language models (LLMs), but it remains unclear whether retrieval uniformly compensates for missing knowledge. We study this question in a controlled factual QA setting over public companies, constructing a benchmark of approximately 2,000 firms across global equity indices. We evaluate six LLMs on four atomic attributes under four conditions: no-context, perfect context, misleading context, and distraction context. We find strong geographic disparities in no-context accuracy, indicating uneven parametric knowledge. While perfect context improves performance, it does not eliminate these gaps: gains are correlated with baseline accuracy, suggesting retrieval effectiveness is coupled to internal representations. Under misleading context, models frequently copy incorrect information. Larger models improve overall performance but do not remove these structural effects. These results challenge the view of RAG as a universal corrective and highlight the interaction between model knowledge, context quality, and entity representation.

Data augmentation is a standard component of modern semantic segmentation pipelines, but most augmentation techniques allocate transformations uniformly across training samples or adapt to a single difficulty signal such as loss. This ignores the fact that segmentation difficulty is multi-factorial, since ambiguous predictions, persistent optimization errors, rare classes, and complex object boundaries can each make a sample informative in different ways. This paper introduces Difficulty-Aware Sample Allocation (DASA), an architecture-agnostic framework that assigns stronger augmentation to samples estimated to be more difficult. DASA combines prediction ambiguity, training loss, class rarity, and boundary complexity into a normalized difficulty score, then maps that score to sample-specific augmentation strength during iterative training. Experiments on Oxford-IIIT Pet and binary Pascal VOC segmentation with U-Net, DeepLabV3, and SegFormer-B0 show that DASA improves over standard training and is competitive with or stronger than single-signal adaptive baselines. On Oxford-IIIT Pet, DASA improves DeepLabV3 from 0.633 to 0.740 mIoU. On binary Pascal VOC, DASA obtains the best foreground IoU for all three evaluated architectures. These results attest to the value of multi-factor difficulty estimation as a practical mechanism for directing augmentation where it is most useful.

We address the simultaneous prediction of multiple high-dimensional physical fields governed by linear equality constraints, a setting that arises in many real-world applications in physics machine learning. Gaussian process (GP) regression is a widely used surrogate modeling approach due to its effectiveness in small-sample regimes and its ability to provide uncertainty quantification. However, applying GP models in this setting raises two major challenges: the high dimensionality of the discretized output fields and the enforcement of the physical constraint in predictions. For the latter, a common strategy consists in deducing one output from the others via the constraint relation. Through a benchmark, we show that this deductive approach is sensitive to the arbitrary choice of which output to deduce, affecting both predictive accuracy and uncertainty quantification. Consequently, there is a need for an approach that treats all fields symmetrically while strictly respecting the underlying physics. Motivated by these limitations, we propose a robust framework for jointly modeling constrained multi-field data. Our approach first leverages a specific PCA procedure for multi-field data, coined row-wise PCA, which has the interesting property of preserving the constraint in the latent space. Since standard PCA strategies for multi-field data do not preserve such constraints, we investigate theoretically the optimality of the row-wise choice. In a second step, we consider a linearly-constrained multi-output GP approach based on a specific kernel parametrization which is trained on the latent space of row-wise PCA. The proposed framework is validated on a population dynamics problem and on an industrial CFD application, which involves the prediction of Reynolds stress tensor components under the incompressibility constraint.

Vision-language models have displayed remarkable capabilities in multi-modal understanding and are increasingly used in critical applications where economic and practical deployment constraints prohibit re-training or fine-tuning. However, these models can also exhibit systematic biases that disproportionately affect protected demographic groups and existing approaches to addressing these biases require extensive model retraining and access to demographic attributes. There is a clear need to develop test-time adaptation (TTA) approaches that improve the fairness characteristics of pretrained models under distributional shift. In this paper, we evaluate how episodic TTA affects fairness in CLIP classification under subpopulation shifts and develop FairTPT, a novel fairness-aware episodic TTA method that jointly minimizes target marginal entropy while maximizing spurious marginal entropy through soft-prompt tuning. We find that standard episodic TTA generally exacerbates disparities between majority and minority groups, that blinding a model to spurious attributes without degrading target performance is inherently challenging, and that excessive blinding can lead to catastrophic forgetting. This model collapse can be prevented by monitoring test-time changes in target loss within the linear regime, while still achieving fairness improvements on reactive data and preserving overall performance. FairTPT outperforms all state-of-the-art episodic test-time debiasing methods and establishes a foundation for robust TTA, which is essential for achieving fairness in practice.

Exact Bayes prediction enjoys fast predictive regret guarantees, but exact posterior updating or representation may be too costly for online use. We study when these statistical guarantees are preserved by computational approximations. We show that the cumulative price of posterior approximation can be governed by the interaction between the contraction radius of the exact Gibbs posterior and the Wasserstein distance between the approximate and exact posteriors. Our general theorem shows that whenever exact Bayes prediction achieves a fast regret bound, any approximate posterior method that tracks the exact posterior with sufficient accuracy inherits the same fast regret, up to an additive term determined by the approximation error. Three online learning examples are developed. For linear models with strongly convex regularized losses, a projected Langevin algorithm yields an approximate posterior that achieves logarithmic regret. For an infinite-dimensional canonical exponential family sequence model over a Sobolev ellipsoid, a prior-preserving truncation method attains the minimax predictive regret rate with sublinear memory and constant update cost per observation. For random-design Gaussian process (GP) regression, a sparse variational posterior with inducing variables achieves the same predictive regret rate as the exact GP, but at substantially lower computational cost.

The Meteosat Third Generation (MTG) Flexible Combined Imager (FCI) offers new opportunities for tropospheric temperature and humidity profiling, at higher spatio-temporal resolutions and expanded spectral coverage relative to its predecessor. Vertically resolved retrievals from broadband imagers are inherently challenging, and operational retrieval algorithms typically rely on numerical weather prediction (NWP) background fields to compensate for limited infrared spectral resolution, reducing the retrievals' independence. We develop a spatially aware deep learning framework to retrieve all-sky tropospheric temperature and humidity profiles from FCI, without forecast profiles as input. A Residual U-Net that exploits spatial context across all 16 FCI channels was trained on 14 months of collocated FCI observations and CERRA reanalysis targets over Europe. Validated against independent radiosondes, retrieved temperatures show biases below 0.4 K and standard deviations of 1.5-1.9 K. Retrieved relative humidity standard deviations range from 12-20 %, compared to 9-19 % for CERRA. Performance degrades modestly under clouds, with standard deviation increases below 0.4 K and 3 % RH beneath cloud tops despite limited direct radiative information. Ablation experiments show that spatial context improves retrievals, with the largest gains below cloud tops. Feature sensitivity analysis indicates broad consistency with FCI bands' established radiative transfer characteristics. Visible and near-infrared channels contribute despite not being commonly used in physics-based profile inversions. These results demonstrate that spatially aware deep learning models can extract statistically reliable tropospheric profiles from geostationary imager observations, independent of NWP forecast fields, enabling more rapid autonomous monitoring of the atmosphere.

Forecasting future measurements from geographically distributed sensors is essential across many domains. However, the spatial distribution of these sensors raises multiple challenges, primarily due to spatial autocorrelation phenomena, that introduce inter-dependencies among nearby locations, that cannot therefore be treated independently. While some existing approaches can capture such phenomena, they generally model the spatial dimension globally across all locations. On the other hand, the method we propose in this paper, called SPALT, focuses on capturing spatial relationships among time series with similar trends, even if they occur at different times, thus modeling the spatio-temporal locality. SPALT leverages linear model trees, which allow us to consider the spatial autocorrelation locally: during the tree-building process, the adopted heuristics group time series exhibiting similar trends into the same node, on which additional features considering the spatial dimension are selectively injected. Additionally, we propose a new pruning strategy, based on Reduced Error Pruning, that also considers the spatio-temporal locality during the tree simplification. Designed for a multi-step setting, SPALT provides forecasts for multiple future time steps across multiple sensors simultaneously. The characteristics exhibited by SPALT can provide significant benefits in different domains, where measurements come from distributed sensors. In this paper, we focus on data produced by sensors located in multiple renewable power plants measuring their energy production at regular, short intervals. Experiments on 3 real-world datasets demonstrate the effectiveness of SPALT in forecasting the production of energy at different time horizons, and its superior performance in comparison with tree-based models and state-of-the-art neural networks that incorporate both temporal and spatial dimensions.

Acquiring pixel-level annotations for medical image segmentation is a severe bottleneck. Traditional U-Net architectures, while effective, learn local texture patterns and lack awareness of global anatomical structures, leading to boundary delineation failures in low-data regimes. This research paper proposes utilizing unsupervised Denoising Diffusion Probabilistic Models (DDPMs) to extract anatomical features. We train a DDPM on 21 unlabeled abdominal CT scans to learn structural representations, transferring the encoder weights to a downstream segmentation task evaluated on the BTCV multi-organ dataset. Diffusion pretraining significantly improved liver segmentation: Dice increased from $0.75\pm0.36$ to $0.93\pm0.16$ ($p < 5.33\times10^{-26}$, 0.529 Cohen's d), Average Surface Distance (ASD) decreased by 66%, and 95th-percentile Hausdorff Distance (HD95) reduced by 45%. For kidney segmentation, Dice improved from $0.90\pm0.19$ to $0.95\pm0.10$ ($p < 4.01\times10^{-11}$). Multi-organ pooled performance showed a 68% variance reduction and a 74% improvement in boundary precision (Dice $0.95\pm0.07$). Crucially, frozen encoder models retained > 80% of fine-tuned performance without exposure to segmentation labels, proving the existence of learned anatomical priors. In low-data scenarios, diffusion-pretrained models maintained robust performance with only 50% (Dice: 0.92 liver, 0.94 kidney), 25%, and even 10% (Dice: 0.89 liver, 0.71 kidney) of labeled data. Using unlabeled images for diffusion-based pretraining successfully embeds robust anatomical features prior to human supervision, transforming U-Nets into anatomy-aware systems.

Predictive models are widely used in many fields, but are vulnerable to false data injection attacks. To address this, detection schemes and adversarial training have been proposed, but such approaches lack guarantees against stealthy attacks. We therefore propose a detector-based switched model, in which optimal attack strategies are stealthy. For linear prediction models, we derive a convex formulation of the resulting adversarial risk. The model incorporates protected features and introduces a hyperparameter modelling attack probability, enabling an explicit performance trade-off between clean and attacked data regimes. Numerical simulations on real and synthetic data show improved performance on partially attacked data, even for misspecified attack probabilities.

Parameter-efficient fine-tuning is often assumed to preserve pretrained capabilities because it updates only a small number of parameters. We show that this assumption depends strongly on adapter capacity. We study factual acquisition in a controlled OpenStreetMap-derived benchmark where Qwen3-4B must acquire anonymized geographic associations while retaining unrelated capabilities. Comparing full fine-tuning (FFT) with quantized low-rank adaptation (QLoRA) at ranks 8, 16, 32, and 64, we find that rank induces a clear acquisition--retention frontier. Low-rank QLoRA preserves out-of-domain (OOD) performance but acquires fewer facts, whereas higher ranks improve same-fact paraphrase generalization at an increasing cost in performance on unrelated benchmarks. FFT behaves as a conservative baseline: it retains general capabilities well, but does not reach the highest factual-acquisition regime. Distributional, weight-space, and spectral diagnostics mirror this behavioral trade-off, with higher-rank QLoRA moving farther from the pretrained model. A separate math adaptation experiment shows a weaker frontier, suggesting that the effect is most pronounced when adaptation must install new factual associations rather than reinforce skills already supported by pretraining. Code and data are available at https://github.com/zhngstl/new_facts_forgetting.

The integration of Large Language Models (LLMs) into cybersecurity has transformed vulnerability assessment, but it has also produced a trustworthiness crisis driven by the unchecked proliferation of "AI slop." These artifacts, hallucinated vulnerabilities, plausible but incorrect patches, and semantically repackaged bug reports, impose a cognitive burden on human triage pipelines that mirrors a denial-of-service attack. This paper surveys the empirical evidence, identifies a unifying mechanism, and traces a path toward trustworthy triage. We formalize a taxonomy of AI slop grounded in a structured literature review and dissect its root cause: the gap between the causal deductive reasoning of security experts and the autoregressive probabilistic generation of current LLMs. We operationalize this gap through a measurable proxy, the Deductive Coverage Score, and show that chain-of-thought prompting and tool-using agents narrow but do not close it. We review mitigation strategies and argue that passive detection and watermarking target provenance rather than correctness, facing fundamental entropy constraints. We instead advocate for active neuro-symbolic verification, mapping each pipeline component to prior systems with documented limits on security inputs. Finally, we specify two evaluation instruments, CVE-Bench and Slop-Score, including dataset construction, metric formulas, and anti-gaming provisions. By shifting evaluation from linguistic fluency to mathematical verifiability, this survey provides a roadmap for securing emerging AI-driven triage systems.

Large language models increasingly mediate access to information, and a growing body of work asks whether they cite the sources behind their outputs. That work treats citation as a verification device and applies it to textual documents. Scholarly citation serves two further functions, credit and provenance, and it applies to data as much as to text. This paper argues that data citation for large language models is an open challenge, distinct from document-level citation grounding and harder to solve. We ask how such models should cite data so that outputs stay verifiable, provenance stays traceable, and credit reaches data creators and curators. We set out three research directions. Training data attribution has to turn influence estimates into references for corpora absorbed into model parameters. Data citation at inference time has to identify datasets, subsets, and query results at the right granularity and fixity. Citing knowledge graph facts has to define what a reference to a single triple denotes and how credit propagates along provenance. Progress on all three depends on joint work across the database, information retrieval, knowledge representation, and artificial intelligence communities.

In 2026, we held the fourth iteration of the SHROOM Shared Task series: SHROOM-Visions (\textbf{S}hared-task on \textbf{H}allucinations and \textbf{R}elated \textbf{O}bservable \textbf{O}vergeneration \textbf{M}istakes in \textbf{Vision} language model\textbf{s}), which is hosted at the UncertaiNLP Workshop co-located with EMNLP 2026. Following the success of the 2024 and 2025 tasks, this time we aim to tackle hallucinations through a model-agnostic detection task focused on large vision-language models. Building on the recently introduced SHEEP dataset, designed for long-term evaluation across model generations, the task invites participants to detect and classify fine-grained hallucination spans in image-conditioned text generation (VQA, image captioning, etc.). The evaluation uses a five-class taxonomy of hallucinations spanning four languages: Chinese, English, French, and Italian. The shared task generated strong interest in the NLP community worldwide, with 27 teams contributing 600+ system submissions. The best systems achieve average scores of 0.58 in character-level correlation, 0.46 in label-conditioned correlation, and 0.51 in intersection-over-union (IoU) across four languages, outperforming the baselines by 30-40 points.

Automated root cause analysis (RCA) with large language models (LLMs) has drawn growing attention. Today, SREs typically automate RCA with LLMs in one of two ways: directly using a general-purpose agent (e.g., Codex or Claude Code) for diagnosis, or building a specialized RCA agent from scratch. As mainstream general agents grow more capable and iterate quickly, our quantitative study finds that the former now often surpasses the latter. Its accuracy, however, still falls short of production needs, and this gap stems mainly from the external adaptation layer outside the agent's general capabilities, namely the harness. We therefore argue that LLM-based RCA should focus on this external harness, reusing the strong general capabilities of a modern agent rather than rebuilding an agent from scratch. A key capability of such a harness is to self-evolve, accumulating system-specific experience from past diagnoses so that it gets better the more it is used. We introduce OpsHarness, a self-evolving RCA harness that turns diagnosis experience into reusable expertise. Its data plane combines layered operational knowledge with an idea-card tool library, while its control plane coordinates setup, diagnosis, evolution, and verification. During evolution, OpsHarness contrasts successful and failed trajectories, converts their evidence into atomic proposals, and admits updates only through a dual-gate verification process designed to prevent overfitting and regression. Across two public benchmarks and an industrial deployment, OpsHarness achieves 59.0\% top-1 accuracy, improving over a bare general agent by 63.4\% and over baseline RCA agents by 4.02$\times$.

Automated novelty judgment can accelerate scientific discovery by enabling efficient evaluation, refinement, and comparison of research ideas. While large language models are increasingly adopted for this task, we investigate a previously overlooked limitation in their judgment capabilities: despite generating reasoning rationales that closely mirror those of human experts, their final novelty judgments often diverge substantially. We demonstrate that this miscalibration stems from a systematic bias towards judging ideas as "medium novel". To mitigate this, we propose Think-Probe-Respond (TPR), a lightweight approach that probes latent novelty judgments from hidden states during the reasoning phase and uses the probed judgments to condition the final response. Across strong baselines, TPR improves novelty judgment performance by 22.30% and successfully mitigates the prevalent "medium novelty" bias.

Large language models (LLMs) excel at programming, but not when the task fixes the target language: prompted with a grammar rare in their training data, their programs usually break the grammar or fail the given specification. Enumerative synthesizers search the space of syntactically correct programs systematically guided by LLMs; the state of the art guides them by approximating LLM proposals into rule frequencies, which loses where each construct belongs and prunes every rule the proposals miss, exactly when the proposals are wrong. We present Narcissus, a synthesizer that keeps the proposals as syntax trees and scores each expansion of a candidate program in its context: does a proposal with the same surrounding structure continue the same way, and does the expansion rebuild a fragment the proposals repeat? A regularization term keeps every rule reachable, so wrong proposals delay the solution but cannot hide it. Across five domains and two search backends, Narcissus beats static guidance at every budget and consistently outperforms re-prompting the LLM to fix its own proposals; it reaches proposal-like programs an order of magnitude sooner and solves $40\%$ of ARC tasks where the raw proposals solve $13\%$, all without a single LLM call during search.

Conversations with chat assistants increasingly span many topics in a single long-running thread, challenging memory systems. Existing long-context and memory benchmarks often expose session or topic boundaries, or probe direct personal-memory questions. These settings understate a harder assistant-memory regime: a flat mixed-topic thread where the system must infer which earlier episode makes a later task decision valid. We introduce SCALE-QA, a constraint-grounded task QA benchmark for flat unsegmented threads targeting episode integrity failure. The dataset contains 3,000 audited questions across 10 domains, uses deterministic four-way multiple-choice grading, and includes a deterministic runtime builder; experiments use all 3,000 questions through 128k and a stratified 400-question diagnostic at 1M. SCALE-QA questions are ordinary task-oriented requests whose correct answer depends on causally related evidence introduced earlier in the conversation. We also propose Temporal-Semantic Interleaved Memory Reconstruction (TSIM), which segments the turn stream into coherent episodes and indexes them through a hierarchical multi-view memory stack with deterministic episode-level summary and cluster-routing views. Experiments show that SCALE-QA challenges strong RAG baselines and long-context LLMs alike; across three open-source and proprietary LLM backends, TSIM achieves the highest accuracy in every backend setting, gaining 5.6-17.6 accuracy points over the strongest corresponding baseline.

Open-ended Theory-of-Mind (ToM) trackers emit valid beliefs absent from finite references. A finite-reference-plus-matcher pipeline marks unmatched outputs false, creating proxy labels that can reverse proper-score model selection on fixed outputs. Holding 259 beliefs and paired scores fixed, reference recoding lowers weighted prevalence from 0.783 to 0.295 and reverses strictly proper Brier risk: a frozen source-prior rule leads native confidence by 0.227 under reference labels and trails by 0.152 under blinded adjudication, in all six authored scenarios. A reference-only Platt recalibrator reverses further. An ICE-specific reversal appears in a released 301-question NQ-open DPR-BERT pipeline: its average-confidence baseline improves instance-level calibration error by 0.045 under exact match but worsens it by 0.074 under human correctness, with both intervals excluding zero. On independently authored OpenToM narratives, 90-96% of audited unmatched beliefs are literally true and the paired direction again reverses. An exact decomposition attributes the distortion to omitted truths, and a closed-form criterion correctly classifies comparisons from twelve released systems. Frozen-audit retrospective replay shows 50 attempted annotations recover ranking direction with probability at least 0.996. TriSource-Restore anchors full-frame reference labels and frozen automatic judgments to a probability-sampled human pilot, maintains at least nominal coverage, narrows intervals, and repairs confidence subject to a base-rate deployment gate.

Portable electronic-nose systems based on low-cost metal-oxide (MOX) gas sensors offer a practical solution for gas and odour recognition, but their signals are affected by slow chemical transients, drifting sensor offsets, scale variation, and cross-channel correlations. Existing pipelines commonly use fixed first-order temporal differencing (FOTD), which requires a manually selected lag and may discard useful response information. We propose LDAC-Net, an end-to-end learnable multi-lag differencing attention-convolution network that operates directly on multi-channel MOX signals. Its learnable differential feature enhancement front-end combines window-conditioned statistical affine normalisation, which compensates for window-specific offset and scale variation, with learnable multi-lag differencing, which weights and combines temporal differences across multiple lags. A compact attention-convolution backbone subsequently models local transients and longer-range temporal dependencies. On the 50-class SmellNet-Base task, LDAC-Net achieves 68.2% top-1 accuracy, exceeding the best FOTD-preprocessed comparison model by approximately 14 percentage points and the raw-input Transformer by more than 30 points. Ablation studies confirm the contributions of both proposed components. The representation also transfers to SmellNet-Mixtures, improving accuracy from 45.4% to 50.5%, and generalises to the 62-channel eNose-Drift benchmark under strong long-term drift, achieving 70.6% top-1 accuracy and 69.6% macro-F1. These results outperform the best comparison model with dataset-retuned FOTD preprocessing by 8.0 and 3.0 points, respectively, demonstrating that learnable, sensor-aware preprocessing is more effective than fixed handcrafted differencing for low-cost MOX gas-sensor recognition.

On-policy distillation (OPD) supervises a student on its own trajectories with token-level signals from a frozen teacher, yet how a sampled loss allocates updates across tokens remains poorly understood. We analyze the gradient of the per-token K2 estimator of reverse KL with respect to the student logits. The $\ell_1$ norm of this gradient factorizes into the absolute teacher--student log-probability gap and a student-side softmax factor that grows as the sampled token becomes less likely under the student. In our math-distillation runs, these per-token norms are highly non-uniform: low-student-probability tokens account for a disproportionate share of their sum and are also enriched in large teacher--student gaps. As a lightweight intervention suggested by this analysis, we study Surprise-aware Reweighting (SuRe), a detached, bounded weighting rule that further amplifies this existing allocation. Across two Qwen3 student scales, SuRe improves several math metrics over vanilla OPD and shows no clear degradation on the selected out-of-domain benchmarks. Our primary contribution is therefore a gradient-level characterization of reverse-KL OPD trained with the K2 estimator, with SuRe as one empirical instantiation.

Traditional task-and-motion planning (TAMP) approaches primarily focus on defining sequences of actions along with the necessary geometric and kinematic constraints to execute long-horizon tasks. However, their applicability in real-world settings is limited, as they typically assume simplified object models that overlook key physical properties critical for the successful execution of contact-rich tasks. Moreover, they often use sub-symbolic reasoning during motion planning, which drastically increases planning time and decreases overall success rates. We propose a method that leverages a TAMP approach, defining object-centric abstractions of execution constraints, called Unified TAMP (U-TAMP), to execute robotic tasks involving interactions among objects with heterogeneous shapes, sizes, and materials. Using a Vision-Language Model (VLM), we generate abstractions of inter-object affordances for characterizing physical interaction constraints between objects in contact-rich tasks, such as grasp and support constraints. These constraints are used to enrich the U-TAMP planning domain to deal with objects with variable physical properties. We perform experiments in simulated kitchen table organization scenarios and compare our results with those of the original U-TAMP, as well as a state-of-the-art VLM-based planner that leverages common sense knowledge of objects' affordances for plan generation. Our approach achieves significantly higher planning success rates and improves planning times by one to two orders of magnitude compared to other methods.

Reference-based verifiers are important for evaluating reasoning models and providing accurate outcome rewards in reinforcement learning with verifiable rewards. To improve verification accuracy, prior work has explored rule-based, model-based, and tool-augmented verifiers for checking answer equivalence across diverse answer forms. However, the equivalence of answer forms such as $1+3.14$ and $1+π$ may depend on the question and scoring criterion. We frame such implicit assumptions as verifier inductive biases. To address this challenge, we propose AutoVerifier, a residual-guided non-parametric optimization method that learns these biases from recurring verifier errors. Specifically, AutoVerifier records these biases in rule cards and promotes them to code modules or prompt guidance only after replay validation detects no direct regressions, keeping accepted updates auditable, editable, and reusable. Experiments on four verifier benchmarks demonstrate that AutoVerifier outperforms state-of-the-art verifiers by a large margin.

Industrial e-commerce search systems ultimately aim to optimize the user-level long-term objective, such as n-day cumulative purchases or gross merchandise value (GMV) per user. However, such objectives are defined at the user level, whereas search ranking is based on item-level scores within each request. Existing methods typically bridge this granularity gap through manually designed multi-objective fusion, where predictions of multiple item-level objectives, such as clicks, carts, purchases, and transaction value, are combined into a ranking score that serves as a proxy for the ultimate objective. Such hand-crafted fusion schemes rely on a small set of manually tuned weights, limiting fine-grained personalization and leading to suboptimal alignment with the ultimate objective. In this paper, we propose DCEO (Direct Causal Effect Optimization), a data-driven framework for learning item-level proxy scores that are better aligned with the ultimate objective. We first aggregate the item-level proxy scores into a user-level proxy metric and quantify its alignment with the ultimate objective using a relative causal effect. We then develop an actor-critic framework, where the critic estimates the ultimate objective for a given user-level proxy metric, and the actor dynamically generates context-dependent fusion weights over multiple objectives to construct the item-level proxy scores and is trained to directly optimize the relative causal effect. Extensive offline experiments and analyses demonstrate the effectiveness and interpretability of DCEO. In addition, DCEO has been deployed in a large-scale industrial e-commerce search system, outperforming the conventional GMV proxy by 0.36% in GMV in a 41-day online A/B test.

Organisations deploying AI face a scoping problem: which tasks can be automated, which should remain with humans, and which are best shared between the two. Aggregate benchmark scores provide little insight into where systems will succeed or fail in practice, while human judgements of model capabilities quickly become outdated. We introduce a pipeline that profiles agents and tasks using a shared set of core cognitive capabilities. Cognitive capability profiling infers an agent's capabilities from performance on a benchmark battery annotated for the cognitive demands of each item. Task requirements weighting elicits from domain experts the relative importance of these same capabilities for their work. As both use a common set of cognitive dimensions, they can be updated independently as models and roles change, and combined to estimate AI suitability at the level of a domain, organisation, role, or individual duty. We validate capability recovery on synthetic agents, profile six AI systems, and elicit task requirements from 410 employees across six occupational domains. AI systems differed more across cognitive dimensions than across model families, while workplace activities converged on a shared cognitive core. The resulting scores provide a comparative scoping tool for identifying promising candidates for piloting and areas where current systems are unlikely to be well suited. We discuss extending the framework to profile human workers alongside AI systems, moving from AI suitability towards human-machine task allocation.

Practical video editing is not only pixel generation: an editor must turn a brief, a clip pool, music metadata, and hard constraints into an executable timeline. We study this decision layer as \emph{executable video-editing planning} and introduce RefineCut, which, unlike workflow systems that wrap a prompted frontier model, trains a compact open-weight planner for it. The planner edits a typed timeline through structured patches covering clip selection, trimming, ordering, transitions, and duration and music alignment; a deterministic verifier applies each patch and checks it against an explicit constraint ledger. Because editing has no single ground-truth repair, we do not imitate teachers directly: RefineCut replays every multi-teacher branch through the verifier and keeps verifier-best repairs as supervision. A second stage, RefineCut-Evo, lets the student score its own repairs with the verifier and a task rubric and trains on high-margin preference pairs, so the final $8$B planner runs in a closed verifier loop with no teacher calls at inference. On RefineCut-Bench ($3{,}578$ tasks, $7{,}971$ captioned clips, $499$ music tracks, explicit ledgers), verifier-replayed distillation lifts the planner from $0.620$ to $0.858$ on the protocol-specific Video-Editing Score and RefineCut-Evo reaches $0.924$; the gain transfers to Llama-3.1-8B and GLM-4-9B, and in the same closed loop the $8$B planner matches or exceeds its frontier teachers. Code and RefineCut-Bench are publicly released; see the Data Availability statement.

Conventional music representations describe acoustic energy over time and frequency but do not explicitly expose relations among simultaneous frequency components. We introduce the \emph{Dissonance Spectrum} (DS), a nonnegative time--frequency representation that applies a tolerance-based rational pitch-relation kernel with logarithmic harmonic distance to a constant-Q spectrum and attributes aggregate pairwise interactions back to individual frequency bins. Controlled music-theory tests show strong ordinal agreement for intervals, harmonic-function connections, and church modes, and weaker but significant agreement across diverse chord voicings. DS is then encoded by a lightweight parallel branch whose zero-initialized residual projection preserves the baseline function at initialization. Across six paired training seeds in open-ended music question answering and categorical and dimensional music emotion recognition, DS obtains the highest mean on every reported endpoint relative to the unchanged baseline, a parameter-matched Gaussian-input branch, and an architecture-matched magnitude-CQT branch. These results support DS as an interpretable, complementary representation, while listener-specific perception and broader task coverage remain open problems.

Long-document visual question answering increasingly relies on VLM agents that retrieve candidate pages, inspect page images, write findings to working memory, and synthesize answers. Working memory should carry answer-supporting evidence across page inspections for later grounded answering, yet existing evaluation mainly checks final-answer correctness and evidence-page access. This creates a memory-quality blind spot: an agent may reach the right page and answer correctly while leaving behind memory too generic or incomplete to support answering once page context is removed. We introduce \emph{memory-only answerability}, a diagnostic that asks whether a reader can answer from the question and terminal working memory alone. Building on this diagnostic, \emph{Answerable Working Memory} (AWM) treats terminal working memory as an answerable evidence artifact, and AWM-GRPO incorporates this signal into the GRPO reward while preserving final-answer priority. Under GRPO, this reward assigns higher advantages to answer-correct trajectories whose terminal working memory remains answerable. On \textsc{MMLongBench-Doc}, even when gold evidence pages are provided, 42.5\% of correct answers still cannot be answered from terminal working memory alone. AWM-GRPO improves final-answer accuracy over the RAG baseline by 8.1 and 11.9 points on \textsc{MMLongBench-Doc} and \textsc{LongDocURL} and reduces the memory-missing-correct rate by 2.7 points over answer-only GRPO.

Large language models (LLMs) demonstrate impressive performance across a wide range of general NLP tasks; however, their effectiveness in sensitive domains, such as hate speech detection, remains less clear. Prior studies comparing prompted LLMs with state-of-the-art encoder-based models (e.g., BERT variants (Roy et al., 2023; Dönmez et al., 2024)) have shown only marginal gains, suggesting that LLMs may not excel in hate speech detection or mitigation. In this work, we revisit this question through the lens of instruction tuning. By thoroughly unifying 36 English hate speech datasets spanning multiple labeling schemes, we fine-tune a generalist LLM, based on Qwen3 (Qwen Team, 2025), specifically for hate speech mitigation. Our results demonstrate not only state-of-the-art performance on in-domain benchmarks but also substantial improvements in cross-domain and cross-lingual generalization--areas where encoder-based specialist classifiers often struggle.

Accurate gust forecasting under typhoon conditions remains challenging due to the highly non-stationary and multi-scale characteristics of extreme wind fluctuations. Existing deep learning models often struggle to simultaneously capture long-term trends and rapid local variations, resulting in degraded performance during extreme events. We propose WDANet, a frequency-aware forecasting framework that integrates stationary wavelet decomposition, a Feature-wise Linear Modulation (FiLM) strategy, and a dual-branch encoder-decoder architecture, enabling separate modeling of trend and fluctuation components. Taking the offshore regions of the Western Pacific in China as an example, we conduct fine-grid wind gust prediction research. The results demonstrate that WDANet shows advantages for short lead times under the experimental setting across a 24-h forecasting horizon and achieves higher prediction accuracy than ECMWF-HRES within the first 6 h. During extreme wind events, WDANet more accurately captures gust peaks and attains the best RMSE and MAE performance. These results highlight its potential for offshore wind power operation, disaster warning, and risk mitigation.

Multi-camera systems are foundational to modern media production, and multi-camera editing is a critical task. This involves the proper selection of the appropriate camera view at each moment. In this paper, we propose a novel Dual-Transformer architecture with Cross-Attention that heavily outperformed the current SOTA models over the TVMCE dataset (TV Shows Multicamera Editing dataset). Our model decouples these tasks: (1) a dedicated temporal encoder first processes the sequence of past frames to build a rich memory of the recent history, and (2) the candidate camera views then act as queries to this memory via a cross-attention module, allowing each candidate to independently interrogate the historical context and find the most relevant information for its own evaluation. Our approach achieved 56.60% Precision@0.5, representing a substantial improvement over the prior best result of 37.16%. We further conducted an ablation study exploring the use of lightweight backbone architectures, where the SwinV2 backbone yielded the best performance, achieving 69.65% Precision@0.5. Using this best-performing configuration, we then investigated the feasibility of adapting the model to replicate the editing style of a specific human editor. To this end, we fine-tuned the model using varying proportions of the initial segment of a target video. Our results demonstrate that even with only 20% of the video used for fine-tuning, the model exhibited measurable improvements in Precision@0.5, indicating strong potential for data-efficient personalization of editing style adapted to each individual TV show or producer.

The purpose of this paper is to provide a general, comprehensive, theoretical framework that allows one to deal with fibrations on graphs labelled on a commutative monoid. This is a genuine extension of the theory of graph fibrations (as introduced in "Fibrations of Graphs" [Discrete Math., vol. 243, pp. 21-66, 2002]), that makes it possible to deal with weighted graphs, and also graphs labelled with other algebraic structures. The derived theory also lends itself naturally to consider approximate fibrations. As an example, we show how this framework can be applied to the compression of arbitrary neural networks (including CNNs), providing a strong theoretical underpinning to the recent results in "The role of fibration symmetries in geometric deep learning" [Proc. Natl. Acad. Sci. USA, vol. 123, no. 4, p. e2416552123, 2026]

Agent capability is not determined by the model alone. The agent harness, encompassing memory management, planning strategy, action protocol, and tool/skill orchestration, can dominate the contribution of the underlying foundation model. Yet harness design remains manual, task-specific, and fundamentally unscalable. We present JIT-Agent, a harness intelligence model trained to synthesize task-adaptive agent harnesses on the fly for arbitrary off-the-shelf agentic LLMs. We formalize the agent harness as a composable, machine-generatable artifact governed by a fixed four-module protocol, and train JIT-Agent to customize harnesses for a given task at hand, repair harnesses for stable and reliable execution, and self-evolve by distilling performance signals from an expanding archive of prior harness configurations. Equipped with JIT-Agent as a harness helper, DeepSeek-V4-Flash surpasses GPT-5.6 on DeepSearchQA (+9.1) and OdysseyBench (+4.3), while the already strong GLM-5.2 gains up to +20.2 points. Across controlled evaluations, JIT-Agent-generated harnesses are performance-competitive with mature agent runtimes such as OpenCode and Claude Code and consistently improve multi-scale model families of DeepSeek V4, Mimo-V2.5, and Qwen3.6. To our knowledge, JIT-Agent is the first model purpose-built for just-in-time harness generation, establishing harness intelligence as a trainable, transferable, and compounding dimension of agent capability orthogonal to model scaling.

Inorganic solid-state electrolytes must combine high room-temperature ionic conductivity, a wide electrochemical window, excellent electronic insulation, and favorable mechanical compliance. Single models struggle to support reliable multi-objective screening across vast chemical spaces because of training-data distribution mismatch, cross-property dataset heterogeneity, and scarce kinetic transport data. To overcome these limitations, we develop a hierarchical synergistic deep-learning framework that sequentially coordinates efficiency, accuracy, and reliability through four complementary modules. The in-house-developed L-G-DCNN and a multi-fidelity implementation built on DenseGNN serve as compositional and structural experts for thermodynamic coarse screening and multi-property evaluation, respectively; MatterSim and system-specific DeePMD models provide transport pre-assessment and kinetic validation. Systematic benchmarks show that each module outperforms mainstream counterparts in its task, while retrospective validation establishes dual closed-loop verification of module-level accuracy and end-to-end workflow reliability. Applied to 30,364,908 Alex/ICSD-derived candidates, the framework identifies 97 high-performance candidates with room-temperature ionic conductivities of 0.109--59.0 mS/cm, including 94 halides, one borohydride, and two oxides. Consistency with independent experimental data confirms that 76 of the 94 halides fall within reported high-conductivity structural regions. Analysis reveals that Li$^{+}$ jump-network connectivity, rather than the number of geometric Li sites, is the core determinant of room-temperature ionic conductivity. Li-defect engineering effectively enhances oxide transport, whereas the inherent rigidity of the O$^{2-}$ framework suggests a potential upper limit on oxide electrolyte performance.

Hierarchical clustering produces ultrametric representations that impose strong global geometric constraints and may distort local similarities in ways that disproportionately affect individual data points. We study hierarchical clustering under an individual fairness requirement that bounds relative distortion within local $k$-nearest neighborhoods. We formulate this requirement as a feasibility problem over dominated ultrametrics and characterize the minimal multiplicative slack required for feasibility. We identify a sharp local threshold, prove stability under bounded perturbations, establish monotonicity in $k$, and show an intrinsic $Θ(\log n)$ separation between local and global realizability. Experiments on synthetic and real world datasets support our theoretical results.

Reasoning-oriented large language models often achieve strong problem-solving performance by generating long chains of thought, but this behavior substantially increases inference cost and latency. In contrast, instruction-tuned models tend to answer more concisely, yet often lack comparable reasoning ability. This accuracy-efficiency mismatch motivates a lightweight approach that combines the strengths of both models without full model retraining. In this paper, we propose GRIP (Granular Reward-guided Interpolation of Parameters), a reward-guided parameter interpolation framework for efficient reasoning. Given a reasoning model and an instruction model with identical architectures, GRIP assigns learnable interpolation ratios to individual modules and optimizes only these ratios while keeping both source models frozen. The interpolation ratios are trained with a reward signal that favors responses that are both correct and concise. Experiments show that GRIP achieves a better accuracy-efficiency trade-off than fixed or search-based merging baselines and further reveals module-wise fusion patterns associated with efficient reasoning.

Concept Bottleneck Models (CBMs) are interpretable-by-design neural networks that detect human-understandable concepts from the input and use them to generate predictions. By allowing users to inspect the concepts underlying a prediction and explore how predictions change under alternative concept configurations, CBMs have emerged as one of the most prominent approaches to supporting human-AI collaboration. However, user studies investigating their actual effectiveness as decision-support systems remain limited. We present two large-scale user studies (N participants = 705, N observations = 6,959) evaluating how concept-based explanations and user interventions on the model's concepts affect the performance of the human-AI team in two distinct binary classification tasks. Our results show that CBMs, and particularly their interactive component, can improve human-AI team accuracy relative to both unaided human performance and performance with non-interpretable AI support. However, these benefits emerge only under certain conditions: classification tasks perceived as difficult, easily identifiable concepts, and active interaction with the model. We also discuss how inaccurate concept detection may undermine users' trust in the model. Overall, this work provides practical guidance for the deployment of CBMs as effective decision-support tools.

Vision-language models can produce fluent answers that are insufficiently grounded in the visual evidence: a single unsupported object, chart value, or intermediate inference can undermine an otherwise plausible response. We argue that this is a credit-assignment failure in multimodal post-training. Scalar outcome rewards indicate whether an answer is acceptable, but do not identify which visual facts are grounded, which reasoning steps are valid, or which instruction constraints are missed. We introduce Visual Rubrics-Based Reinforcement Learning, which decomposes reference responses into atomic propositions and scores generated answers along Visual Faithfulness (VF), Reasoning Consistency (RC), and Instruction Following (IF). The resulting rubric items provide structured partial credit and localize rubric credit when supporting evidence spans are available. We first obtain an SFT checkpoint by fine-tuning Qwen3-VL-8B-Instruct on the public OpenMMReasoner-SFT-874K corpus, adapting OpenMMReasoner's cold-start data recipe. We construct V-Rubrics 50K, a 50,248-example training set from 17 visually grounded sources, by applying rule-based filters before deriving example difficulty from rejection-sampling scores and then annotating every example with Gemini-3-Pro under the same structured prompt and protocol. We train our model based on the same SFT checkpoint using component-wise, prefix-localized rubric credit. Experiments show that our rubricbased GRPO improves over both the shared SFT baseline and answer-only GRPO, with the largest gains on knowledge-oriented and visually grounded reasoning benchmarks. The results show rubrics as a useful reward abstraction for visual post-training.

Metaphor-identification performance can change markedly across datasets that differ in text distribution and annotation policy. We examine whether a fixed expert-informed procedure produces a more even cross-dataset profile than task-specific parameter adaptation. Four prespecified conditions are compared for Chinese sentence-level metaphor identification: BERT fine-tuning (BERT-FT), QLoRA-based large language model fine-tuning (LLM-FT), direct zero-shot LLM prompting (LLM-ZS), and zero-shot prompting with a frozen procedural Skill (Skill-ZS). The Skill operationalizes established criteria involving contextual meaning, basic meaning, contrast, and comparison. Evaluation covers CMRE Test and two external datasets, CCIME and CMC. Fine-tuned scores are means over three seeds, whereas each zero-shot score comes from one deterministic configuration. Fine-tuning remains strongest on the native test set: BERT-FT reaches 91.76 Macro-F1. LLM-FT has the highest external mean (83.52), while Skill-ZS is close at 82.92 and has both the highest external floor (82.64) and the smallest observed range across all three datasets (4.08 points). In the matched zero-shot comparison, adding the Skill reduces metaphorical predictions on every dataset. This sharply lowers false positives on CCIME but increases false negatives on CMRE Test and CMC. The results position expert-informed Skill prompting as a complementary route to more even observed cross-dataset performance, while fine-tuning retains its advantage in native-data accuracy. To our knowledge, this is the first study to compare an expert-informed procedural Skill with task-specific fine-tuning in the same cross-dataset evaluation of Chinese sentence-level metaphor identification.

With the widespread adoption of personal intelligent agents, users generate massive, heterogeneous data during long-term interactions. Leveraging this data as long-term memory helps reduce token overhead and deliver personalized experiences. However, existing memory systems face two primary limitations: they rely on single-storage paradigms that fragment multi-dimensional data, and they lack fine-grained data provenance to resolve long-term factual conflicts, thereby worsening LLM hallucinations. In this demonstration, we introduce PolyMemDB, a novel system tailored for managing agent memory. PolyMemDB has a polyglot storage architecture designed to track and manage various memory types, including graph, vector, probability and spatial-temporal data. To ensure factual consistency and reduce hallucinations, it features a probabilistic inference engine that integrates temporal decay with semiring aggregation, resolving long-term factual conflicts, providing detailed data provenance, and enabling users to trace reasoning chains transparently.

Pretrained vision-language models such as CLIP excel at zero-shot recognition but often fail at compositionality, particularly attribute-object and relational structures. Recent studies mitigate this issue by augmenting training with synthetic hard negatives generated by a cascade of large language models and text-to-image models, which incurs substantial pipeline overhead. We instead propose MLLMCLIP, a heterogeneous distillation framework that transfers multimodal knowledge directly from a generative Multimodal Large Language Model (MLLM) teacher into a discriminative CLIP student, bypassing synthetic data entirely. To bridge the architectural mismatch between the two paradigms, we introduce an attention-based per-layer token selection and a CKA-based distillation loss. Compared to prior CLIP-enhancement methods, MLLMCLIP achieves state-of-the-art compositional accuracy while delivering consistent gains on standard zero-shot classification and image-text retrieval, showing that feature-level distillation strengthens both compositional and general vision-language representation capability.

Automatic Speech Recognition (ASR) is typically evaluated using Word Error Rate (WER), which poorly reflects semantic similarity. While embedding-based metrics correlate better with human judgments, the respective roles of encoder and decoder-based Large Language Models (LLMs) remain underexplored. This paper presents a comparative study of both families for ASR evaluation. We analyze BERTScore and SemDist across different LLMs, layers, and pooling strategies, showing that both metrics can achieve strong correlation with human judgments when properly configured. For decoder models, we investigate generative LLMs in two settings: pairwise hypothesis selection via prompting and direct qualitative error classification. Our results show that encoder-based metrics remain highly competitive, while generative LLMs perform strongly in hypothesis comparison and improve the interpretability of ASR evaluation.

Relational Database Management Systems (RDBMSs) are the backbone of modern data-intensive applications, making reliability and robustness critical. However, achieving high coverage in RDBMS testing remains challenging because of large codebases and complex execution logic. Traditional fuzzing relies on random SQL generation and cannot capture the correspondence between SQL inputs and internal execution paths, while symbolic execution suffers from prohibitive cost and scalability limitations. We propose DBcover, an LLM-driven white-box SQL test generation framework based on contextual reasoning. DBcover uses lightweight dynamic analysis to extract SQL-to-path correspondence and call graphs as global context, and collects source-level information around target functions as local context. These contexts are organized in a unified knowledge graph for efficient retrieval and reuse. DBcover then performs two-phase test generation: it first selects a semantically relevant seed whose execution path is close to the uncovered target, and then guides the LLM with global and local context to generate SQL test cases that trigger previously uncovered code regions. Experiments show that DBcover achieves 80.1% and 82.3% coverage on PostgreSQL and MySQL, and is also effective on the enterprise RDBMS KingbaseES, demonstrating its practical applicability to closed-source systems.

Action-conditioned world models have become an important foundation for embodied prediction, planning, and synthetic data generation, but their errors under new task and scene distributions are often concentrated in localized spatiotemporal regions such as robot arms, manipulated objects, contact areas, and occluded objects. This paper presents ConfAL-WM, a confidence-guided active learning framework for post-training embodied world models. Built upon EVAC, we attach a lightweight confidence probe to UNet decoder features and predict dense confidence maps in the latent space. These maps are aggregated into task-, frame-, and patch-level scores, enabling both efficient data selection and localized training enhancement. Our pipeline first retrains the confidence probe and warms up EVAC with a small subset of target-domain data, then performs task-level prescreening to allocate sampling budgets, and finally applies selected-data retraining with optional frame or patch weighted data enhancement. Experiments on RoboTwin2.0 show that confidence-guided selection improves post-training efficiency, while dense frame and patch weighting further enhances prediction quality and embodied trajectory consistency compared with scalar reward, progress, and judge-based scoring baselines. A quick visual overview of this work is available at https://ConfAL-WM.github.io.

Hardware kernel optimization requires repeated compilation, correctness testing, profiling, and revision. LLM agents can automate parts of this process, and stronger foundation models, longer context windows, and longer execution horizons have improved optimization within individual tasks. These advances alone do not enable an agent to learn from completed optimization runs. Existing kernel-optimization agents seldom preserve a decision, its observed execution feedback, and the later decisions that use that evidence. Retaining every prior trajectory is also impractical because an expanding history competes with the current task for context. We present KOPE, an experience-driven framework for hardware kernel optimization. KOPE records optimization trajectories with correctness and performance feedback in Experience Graph Memory, then uses Active Context Management and Injection to retrieve relevant experience under a fixed token budget. The graph retains decision order, observed outcomes, and alternative branches, allowing evidence collected on the target hardware to inform later optimization steps and tasks. Under the same GLM-5.2 setting, the geometric mean of KOPE's per-operator speedups is $1.54\times$ that of CANNBot, the strongest competing baseline. In a complete 53-operator ablation, Active Context Management and Injection raises pass rate from 60.0\% to 84.6\%, increases the evaluator-reported positive-field geometric mean from 0.0382 to 0.0661, and reduces optimization token consumption from 15.9B to 1.113B tokens relative to passive agent-led context construction. Enabling Experience Graph Memory raises full-suite pass rate from 55.2\% to 84.6\% and yields a $1.43\times$ geometric-mean speedup on valid timing comparisons. These results support continual optimization through external experience while the foundation model remains fixed.

Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further refines them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectiveness of vector purification and adaptive scaling, establishing EmoVec as a practical inference-time method for affective control in deployed LLMs.

To reduce deployment cost and retraining overhead, adapting pretrained learned image compression (LIC) models to downstream machine vision tasks has attracted growing attention. However, existing methods typically insert fine-tuning modules independently into frozen backbones, lacking explicit mechanisms for cross-layer coordination. To address this limitation, we propose a novel framework named CrossMambaTuning, which integrates State Space Models with cross-layer interaction mechanisms for parameter-efficient fine-tuning. Specifically, we design an efficient Mamba adapter equipped with task-specific prompts and multi-scale branching to precisely capture both local features and global dependencies. Furthermore, we introduce a Scale-Invariant Cross-Layer Adapter (SICA) utilizing a parameter-sharing strategy to fuse task information across different scales and reduce redundancy. Extensive experiments demonstrate that CrossMambaTuning achieves state-of-the-art (SOTA) performance on multiple machine vision tasks, reducing parameter overhead by 72\% compared to SOTA methods. Code is available at https://github.com/rsr1123/CrossMambaTuning.

Physics-informed neural networks (PINNs) often rely on over-parameterized models to optimize coupled solution and differential-residual objectives, leaving unclear how much capacity is necessary and what pruning should preserve. We study foresight pruning at initialization for sparse PirateNet PDE solvers. Standard neural tangent kernel spectrum-aware pruning (NTK-SAP) aims to preserve output-side training dynamics but may overlook parameters whose main influence arises through derivatives in the governing equations. We introduce physics-informed spectrum-aware pruning (PI-SAP), which assigns saliency using sensitivity of the PDE residual. Experiments on the Gray-Scott equations, complex Ginzburg-Landau equation, Burgers' equation, and linear convection equation show that PI-SAP more consistently preserves Gray-Scott residual fidelity and is competitive under aggressive sparsity. However, no criterion is uniformly optimal across equations or sparsity levels. Small-batch PINN-NTK diagnostics further show that residual fidelity, solution accuracy, and kernel conditioning are distinct objectives, motivating pruning methods that explicitly balance solution-side and residual-side training dynamics during optimization.

VLMs are increasingly positioned as daily assistants that perceive first-person environments, follow user dialogue, and decide how to help. Existing egocentric benchmarks mainly evaluate visual understanding in isolation, leaving open whether models can arbitrate between visual evidence and user-provided language when the two are helpful, irrelevant, or conflicting. We introduce EgoArgus, a human-annotated dataset for evaluating egocentric assistants on understanding and decision tasks in five dialogue-video daily scenarios. Our results demonstrate that it is still challenging for current VLMs as reliable egocentric assistants, which requires identifying which modality is trustworthy and deciding when intervention is warranted. Deeper analysis also shows that existing modality bias mitigation methods are quite restricted to enhance performance, providing insights to aid practioners into the deployment of current VLMs as daily assistants.

Video deep research answers complex questions by jointly understanding video content and retrieving external knowledge from the open Web. However, diverse questions and videos require different tool-use strategies, and inappropriate tool calls can produce incorrect results. Uncertain grounding and retrieval also make unnecessary interactions costly and error-prone, increasing latency and reasoning errors. To address these challenges, we propose AdaVDR, an adaptive video deep research agent with adaptive tool invocation and reflection. AdaVDR selects tools according to the task and its capabilities, and backtracks only when unreliable intermediate results require correction. To enable these capabilities, we develop a video deep research data construction pipeline. We first discover retrieval-relevant events and entities in diverse videos and acquire detailed information through grounding and external retrieval to construct high-quality QA pairs. For each QA, task-specific prompts organize the information acquisition process into a tool-use trajectory, allowing different question and video types to follow different grounding and retrieval strategies. We further introduce model-conditioned tool necessity filtering, which evaluates tool calls against the target model's video understanding and internal knowledge, removing tools or tool chains the model can bypass. This yields trajectories tailored to the target model's video understanding capability and knowledge. Using this pipeline, we construct training data and VDR-EE, a benchmark covering entity-centric and event-centric questions. We perform supervised fine-tuning followed by reinforcement learning with a redundancy-aware reward to strengthen adaptive tool invocation and reflection. Experiments show that our method performs best among the evaluated open-source models on VDR-EE and substantially improves over its base models on VideoDR.

Explainability for transformer-based language models is becoming crucial as these systems are deployed in high-stakes applications. As a result, the ecosystem of explainability tools is rapidly evolving, becoming richer, but also more fragmented and harder to navigate. To address this challenge, we present Virgil, an interactive system that lets practitioners and researchers, including non-experts, navigate explainability tools for transformer language models. Supported by a curated knowledge base, the system enables users to discover and compare explainability tools within a unified interface.

An agent that inherits a consolidated memory may inherit a constraint that was true when written and has since been withdrawn by a newer authoritative record. Under a scarce verification budget, does the agent recover the withdrawal, and if not, is the error avoidable without spending more? We model supersession explicitly -- historical provenance is immutable; what changes is which record is current -- and assign by design the memory's form, the world's state (source current or superseded), and the verification policy at a fixed budget of two records: the agent's own allocation, or the same budget with one slot re-assigned to the critical provenance path or to a random record. With a constraint stated, agents inspected its provenance path in about one episode in five; when that constraint had been superseded, native allocation produced stale-consistent decisions in 77.3%, 74.7% and 74.7% of episodes across a primary run, a fresh-wording replication and a held-out domain. Re-assigning one slot to the critical path raised current-record-consistent decisions by +74.0, +72.7 and +61.3 points, positive in six of six models in each of those runs, and changed nothing when the record agreed with the memory. The held-out scenario was later found to contain a temporal inconsistency; a robustness replication with one sentence corrected, deposited externally before execution, gave +73.3 points and is reported alongside the original. The intervention uses knowledge of the critical path and is not a scheduler; it identifies that the share of stale-memory error attributable to verification allocation is close to its structural ceiling. Memory systems may need freshness or supersession signals separate from relevance.

Stochastic gradient descent (SGD) is typically analyzed at a deterministic horizon chosen before the algorithm is run, even though practical stopping decisions are made adaptively by inspecting the evolving trajectory. This mismatch creates a fundamental certification problem: fixed-time guarantees do not generally remain valid at data-dependent stopping times, while deterministic horizons derived from worst-case bounds can be highly conservative. We address this problem for strongly convex stochastic optimization by constructing fully observable, trajectory-adaptive upper confidence sequences for the squared distance of the last iterate to the optimizer and the suboptimality of a weighted average. These bounds hold simultaneously over time, attain the optimal $1/t$ decay rate up to iterated-logarithmic factors in the worst case, and adapt to the realized stochastic gradients, allowing SGD to stop as soon as a prescribed accuracy is certified without sacrificing statistical validity. Our approach treats the evolving SGD trajectory as a sequential experiment whose observations provide evidence about the unknown optimization error. To formalize this perspective, we develop new recursive confidence-sequence techniques and a general time-uniform empirical Bernstein inequality for adapted processes with time-varying conditional means and predictable ranges that may grow without bound. We further extend these confidence-sequence constructions to minibatch SGD, with the empirical Bernstein bounds exploiting the realized second-moment structure within each minibatch. Numerical experiments show that the resulting stopping rules can require several orders of magnitude fewer iterations than natural deterministic horizons.

The success of any mobile application relies on its usefulness and rating is considered as an important measure in this regard. This research work focuses on identifying usability factors, which contribute significantly towards the rating of M-commerce apps. This work intends to explore existing usability models consisting of different factors along with a set of criteria and evaluate in terms of rating estimation by considering 5 well-known mobile applications, namely (i) daraz, (ii) shophive, (iii) home shopping, (iv) Symbios. (v) yayvo. Then, this work provides a hybrid usability model for rating prediction of M-commerce applications. The initial hybrid usability model comprises of (i) learnability, (ii) consistency, (iii) human factors,(iv)communicativeness,(v)effectiveness, (vi) Operability, (vii) efficiency, (viii) satisfaction. Each factor consists of some criteria. Keeping in view the factors of hybrid usability model, the data was collected from 40 users for each application. Furthermore, Forward Stepwise Multiple Linear Regression based rating prediction model is suggested by analyzing each criterion of all factors of hybrid usability model. Finally, the model is assessed and validated by using PRED(x) and K-fold techniques.

Recently, there has been a growing adoption of protein language models (PLMs) in biomedical science. Their embeddings provide a rich numerical representation of protein sequences which achieve state-of-the-art performance on several downstream tasks including protein fitness prediction. However, PLM embeddings are not directly interpretable and, thereby, it remains unclear what features they encode. To gain insight into which biochemical properties of the protein are driving the prediction, we leverage an orthogonal projection technique that removes linear effects of known tabular features from embeddings and extend it to high-order and interaction effects. In this way, we remove the effects of interpretable biochemical features from PLM embeddings. In an ablation study, we show that this leads to a decrease in performance for a downstream classifier trained only on the embeddings to predict protein fitness. In an additional evaluation, we find that these biochemical features explain a substantial part of the variance in the predictions of this classifier. Hence, we can show that PLM embeddings encode patterns correlated with biochemical properties and quantify their contribution to predicting protein fitness. This computationally efficient approach is not limited to the features or embeddings considered here and is readily transferable to problem settings beyond protein fitness prediction.

To apply the techniques of embodied artificial intelligence to human-oid robots for complex manipulations, dexterous robotic hands are indispensable, which are restricted by the dexterity and tactile perception capability. In this work, we proposed a novel design of tendon-driven five-fingered hand with dis-tributed tactile perception. With a soft-rigid-hybrid structure employed, both compliance and operational force are endowed to the hand. Dual-modality tactile sensing elements are distributed on the distal and middle phalanges of all five fingers, enabling the simultaneous detection of static contact and dynamic force variations. Manipulation experiments, including counting gestures, finger-to-thumb pinching, object grasping, and bottle-grasp tactile recording, demonstrate the feasibility of the integrated actuation-perception system.

This paper presents a novel two-phase maskable proximal policy optimization (TP-MPPO) algorithm, which maximizes the system goodput counting request throughput with strict service level objective (SLO) compliance for large language model (LLM) inference services in wireless edge networks. In the first phase of TP-MPPO, we optimize the task offloading decisions by MPPO with action masking mechanism, effectively avoiding exploring invalid actions and reducing the action space. In the second phase, closed-form solutions are derived for uplink bandwidth allocation; a greedy algorithm is designed for downlink bandwidth allocation to provide immediate rewards for the MPPO in the next round. The two stages alternate till convergence. Simulation results demonstrate that TP-MPPO can improve the system reward by 33.3%--87.5% compared to its benchmarks and achieve the highest goodput.

Large reasoning models often produce reasoning traces with verification, revision, and backtracking. When reflection merely re-checks established results, it wastes reasoning tokens and increases latency. Most existing reflection steering methods add a label-derived mean-difference direction across preset layers, but its entanglement with reasoning and length signals destabilizes the accuracy-efficiency trade-off. In this paper, we propose Reflection Steering, a training-free framework for controlling reflection-associated computation within LLMs by disentangling reflection-related activations from general reasoning. Specifically, we contrast reflective and non-reflective hidden states at each LLM layer, denoise the resulting reflection directions with PCA, and orthogonalize them against general-reasoning directions. To limit downstream amplification from early-layer interventions, we calibrate each layer across multiple intervention strengths on a small set, retain only stable layers, and apply bounded projection removal to their residual-stream activations. We conduct extensive experiments across two public benchmarks and three open-weight LLMs against state-of-the-art activation-steering baselines. Results show that Reflection Steering reduces reasoning tokens by 16.9% on average across six matched settings. Besides, our method further introduces a bounded reflection intervention-strength parameter $α$, enabling deployment-time adjustment to balance token savings, accuracy, and generation stability.

A plant-health score can appear precise while resting on duplicated image families, a long-tailed label space, or a runtime file that was never evaluated. We present CropCop, a closed-set recognition system spanning 120 operational plant-health classes and an evidence chain from corpus reconstruction to direct execution of the final quantised artifact. Starting from 117,546 audited images, we rejected the inherited partition after confirming 3,233 duplicate relationships across split boundaries and froze a 109,107-image benchmark with zero crossings among the audited trusted leakage groups and a 151.7 largest-to-smallest class ratio. A fully fine-tuned DINOv3 ConvNeXt-Tiny reference achieved 98.51% accuracy and 96.87% macro-F1 on the locked internal test. A compact MobileNetV4 Conv-Medium derivative achieved 98.46% accuracy and 96.27% macro-F1 without being presented as evidence for a new distillation method. Validation-only post-training quantisation selected dynamic activations with per-channel weights, and the final 22.60 MiB ExecuTorch/XNNPACK PTE achieved 98.46% accuracy and 96.23% macro-F1 when executed directly. Only six of 16,363 top-1 decisions changed between the converted INT8 graph and the PTE, while paired analysis showed a modest class-balanced loss; an exploratory post hoc fruit-label slice localized a larger recall decline than aggregate accuracy revealed. CropCop establishes strong leakage-controlled internal recognition and software-runtime fidelity; it does not establish performance on unseen farms, camera pipelines, or physical Android hardware.

In this work, we propose a view on electronic Neuromorphic Design Automation (eNDA), which we see as a design automation flow that bridges computational neuroscience modeling with traditional Electronic Design Automation (EDA) flow. We introduce the term, give examples of how it can be implemented, and design a prototype implementation: Syn2Logic. Syn2Logic is an entire eNDA framework, that allows neuroscientists to model neural behavior using a custom DSL and a compiler that takes the same model description down to synthesizable RTL hardware. We end the paper by applying the eNDA-flow through Syn2Logic to show how to -- without writing a single line of hardware description language (HDL) code-- (i) generate what we believe is the fastest C. elegans accelerator that runs significantly faster than state-of-the-art simulators, (ii) create (to the best of our knowledge) the fastest, most generic neuromorphic sudoku solver that outperforms CP-SAT and SCIP on TOP1465 puzzles, and (iii) create a 5.6 million FPS/Watt accelerator on a tiny FPGA that outperforms existing neuromorphic architectures in terms of speed and energy-efficiency on the MNIST dataset.

Wireless Internet-of-Things (IoT) edge networks require decentralized learning (DecL) methods that can operate reliably under both heterogeneous local data and communication-constrained wireless links. However, existing decentralized optimization schemes often incur substantial communication overhead and degraded performance when transmissions are constrained by strict airtime budgets, fading channels, and packet losses. This paper proposes QEF-GT-AdamW, a communication-efficient and outage-resilient algorithm for DecL over wireless communication (WCom) networks. The proposed method combines gradient tracking to mitigate the effect of non-IID data, AdamW-based adaptive optimization to improve training stability, and dual-stream biased quantization with error feedback to reduce communication payloads for both model and tracking exchanges. To address unreliable broadcast communication, the proposed framework further employs a local fallback strategy when scheduled packets are not successfully received. We explicitly model the effect of bandwidth, transmit power, airtime constraints, and fading channels on DecL performance, and establish convergence guarantees for the proposed algorithm under compressed and unreliable wireless communication. Experimental results on heterogeneous MNIST and CIFAR-10 settings show that QEF-GT-AdamW consistently improves robustness and convergence performance over representative DecL baselines while achieving favorable accuracy-communication trade-offs under limited wireless resources.

Humanities and social science research requires close reading of long narrative materials such as novels, scripts, archives, and case reports, yet many users have limited access to costly proprietary long-context models. Compact, locally deployable language models are a practical alternative, but directly feeding them an entire long context remains costly, hard to inspect, and prone to missing sparse evidence. We present ClueWeaver, an evidence-aware dual-agent framework for long-narrative question answering with compact local models. A Finder identifies passages containing answer-critical clues through retrieval-guided segmentation, while an Interpreter derives the answer from the selected evidence, produces rationales with paragraph-ID citations, and applies an internal self-calibration pass for high-risk questions. Both agents are optimized with reward-guided reinforcement learning: Finder rewards emphasize evidence retention and faithful paragraph-ID references, and Interpreter rewards emphasize correctness, grounding, and concise explanations. This decomposition makes evidence selection and reasoning more inspectable than end-to-end prompting. Experiments across multiple long-context narrative question answering and claim verification settings show that ClueWeaver substantially improves local end-to-end language models while providing evidence coverage and paragraph-referenced reasoning traces. Code is available at https://github.com/Ameame1/ClueWeaver.

The Levenberg-Marquardt (LM) algorithm is the most widely used method for solving nonlinear least-squares problems, as it combines the robustness of steepest descent with the fast local convergence of the Gauss-Newton method. However, its computational cost can become prohibitive for large-scale problems because each iteration requires solving a large damped linear system, and conventional step acceptance strategies may require repeated solves as the damping parameter is adjusted. Despite this computational challenge, many large-scale least-squares problems exhibit effective low-dimensional structure, with only a small number of parameter-space directions strongly informed by the data. We propose an adaptive hybrid subspace Levenberg-Marquardt (HSLM) algorithm that constructs a low-dimensional subspace from complementary sources of gradient, memory, Krylov-subspace, and randomized curvature information and computes a spectrally damped LM step within this subspace. A distinguishing feature of the method is a deterministic adequacy monitor that quantifies how much descent information is captured by the reduced space and adaptively enriches the subspace when necessary. Step acceptance is decoupled from damping adjustment: Armijo backtracking determines the accepted step length, while the ratio of actual to predicted reduction is used solely to update the damping parameter, thereby avoiding repeated damped-system solves during step acceptance. For the HSLM algorithm, we establish global convergence to stationarity and prove local linear and superlinear convergence. Numerical experiments on neural-network training problems show that HSLM achieves convergence behavior comparable to classical and Krylov subspace LM (KSLM) while substantially reducing per-iteration computational cost, with increasing advantages observed as the parameter dimension grows.

Prefix caching introduces a fundamental tradeoff in multi-agent large language model (LLM) serving: retaining a long system-prompt key-value (KV) cache for an agent accelerates future calls, yet it reduces the GPU memory available for batching concurrent requests. In multi-stage workflows, existing schedulers tend to prioritize either immediate prefix locality or overall workflow progress. However, under a shared KV cache budget, optimizing either objective in isolation can prolong tasklevel job completion time (JCT) through downstream delays or frequent prefix replacement. To strike a balance, we here propose TOPAS, a Task-Oriented Prefix-Aware Scheduler that jointly decides which agent prefixes to keep in the cache and which requests to schedule for execution. TOPAS scores candidate post-decision states by trading off the expected reduction in each task's longest remaining service path against the near-term benefit of downstream prefix reuse, accounting for the costs of prefix movement and preemption. A task-level aging mechanism is also incorporated to prevent starvation. We implement TOPAS within the SGLang framework and assess its performance on three synthetic DAGs and two MetaGPT software-development workflows. Compared with the best performing baseline for each workload and metric, TOPAS reduces the mean/p99 JCT by up to 39.8%/49.4% on the synthetic workloads, while lowering mean JCT by 9.8% on MetaGPT-SOP and mean/p99 JCT by 22.0%/26.6% on MetaGPT-TL.

A common strategy for scaling world models is to train on more crawled video with more compute. We argue that this strategy is inefficient: scaling world models also requires a recursive data engine that offers grounded reward signals. The success of code agents illustrates why this matters. As code is executable, compilers and runtimes can provide high-quality rewards for Reinforcement Learning (RL) post-training of LLMs. By contrast, spatial generation still relies largely on fuzzy proxies such as CLIP scores. These signals are fuzzy and biased, making them hard to support RL post-training. Compared with these, game development provides a missing reward environment for spatial world models. A scene encoded by a game engine is an executable world specification: the engine can efficiently check collision, physics, navigability and bounded playability, while the developer provides the global verification signal by judging whether the scene should be accepted. Game development also provides real-world long-horizon trajectory data for RL post-training. We therefore propose Reinforcement Learning with Human-Engine Verification (RLHEV), a post-training paradigm that combines dense engine signals with implicit human acceptance feedback from the development process.

Random feature methods provide a scalable approximation to kernel ridge regression (KRR), but the regularization parameter that yields the oracle learning rate depends on unknown smoothness and capacity parameters. In this work, we propose a neighboring early-stopping rule for adaptive regularization in KRR with random features (KRR-RF). The method uses a grid that is uniform in inverse regularization and compares only adjacent estimators, reducing the number of discrepancy comparisons relative to standard all-pairs Lepskii-type procedures. Both the neighboring discrepancy and its empirical complexity term can be computed directly in the random feature space, without constructing the exact kernel Gram matrix. We establish a high-probability comparison bound for neighboring KRR-RF estimators and show that, under standard source and capacity conditions together with suitable grid and random feature budget conditions, the selected estimator attains the oracle polynomial learning rate up to logarithmic factors. The result allows the regularization parameter to be selected without prior knowledge of the source and capacity exponents and covers both well-specified and partially misspecified regimes. Our analysis is based on an empirical random feature effective dimension that connects the observable stopping threshold with the population complexity of the random feature model. Simulation and real-data experiments illustrate the prediction performance and computational behavior of the proposed method in comparison with standard tuning procedures.

Modern software -- from plugin systems to self-evolving agent harnesses -- increasingly requires dynamic composition, yet its formal foundations remain underdeveloped. We identify two orthogonal dimensions of the problem: temporal composability, the ability to completely revert a component's side effects upon removal, and spatial composability, the ability to declare and reactively manage inter-component dependencies. We address the two dimensions by lifting classical effect and coeffect concepts to runtime mechanisms. In particular, we formalize revertible effects, in which every context transformation carries an inverse that the runtime holds, establishing temporal composability local to one component. We formalize reactive coeffects, in which every context change is classified against a component's coeffect specification to drive its activation and deactivation, establishing spatial composability local to one component. We then unify the effect context and the coeffect context into a single context type and mediate every effect and coeffect through it, yielding a discipline we call the context paradigm; the mediation induces an observational equivalence up to which the effects of distinct components interleave without disturbing one another. Combining these mechanisms into the notion of a component, we give a calculus of dynamic composition whose metatheory carries spatiotemporal composability from a single component to a whole system of interleaved components. We implement these ideas in Cordis, a meta-framework of spatiotemporal composability that provides a core library with effect tracking and coeffect resolution, as well as a declarative component loader with configuration reconciliation and hot module replacement.

Motivated by parallel decoding in masked diffusion models, we study adaptive parallel sampling of discrete vectors: in each round, a deterministic policy selects unrevealed coordinates on the basis of the values observed so far, and the selected coordinates are sampled independently from their exact conditional marginals. Approximation error is measured by forward Kullback-Leibler divergence, and serial depth is the minimum target-averaged number of rounds meeting a prescribed error budget. Our central result is an exact identity: the divergence of every policy equals the expected conditional total correlation accumulated over its reveal rounds, so conditional total correlation is the exact information cost of within-round parallelism. The identity yields zero-error schedules for finite-order Markov chains with round complexity proportional to the Markov order and logarithmic in sequence length, a matching logarithmic characterization of the Bernoulli walk at every fixed error budget, and a linear-versus-logarithmic separation between left-to-right and hierarchical reveal orders. Uniform random permutations require linearly many expected rounds at every fixed budget; their hard-cap round-error tradeoff is an exact integer-composition problem whose fixed-round asymptotics and joint-scaling frontier we determine. Uniform balanced binary strings have depth of order squared logarithm, and binary one-hot blocks have square-root depth, with rectangular versions realizing every polynomial exponent up to one half. These results separate serial depth from entropy and negative log-likelihood, and establish conditional-dependence structure as a fundamental determinant of parallelizability. Experiments with a masked diffusion language model show that the pseudo-cost distinguishes deployed decoding rules and that its policy rankings agree closely with the quality of self-sampled outputs.

Reusable skill libraries allow large language model (LLM) agents to reuse procedural knowledge across tasks, but they also turn memory access into a challenging retrieval problem. Full-library prompting preserves coverage at high context cost, vector retrieval returns compact neighborhoods but treats skills as independent text, and graph-based retrieval can recover workflow context only when the edges that carry relevance are reliable. We propose CaSKG, a counterfactual-causal skill graph framework that calibrates procedural relations before retrieval. CaSKG first builds a high-recall directed candidate graph from semantic, lexical, input/output, and structural evidence, with repair evidence and an optional LLM judge further refining candidate scores. It then applies direction-conditioned textual counterfactual probes that remove, substitute, and reorder skill pairs, aggregates the evidence with Bayesian smoothing, and publishes a state-filtered weighted graph for task-conditioned expansion. The graph is constructed offline and used without changing the downstream agent policy or task interface. Across six LLM backbones on ALFWorld ID-140 and ScienceWorld U211, CaSKG achieves the highest task score in all twelve combinations of model and benchmark. Relative to Graph-of-Skills (GoS), it improves the six-model macro-average ScienceWorld score from 72.62 to 80.50 and ALFWorld success from 80.01\% to 86.79\%, while reducing mean environment steps on both benchmarks. Qualitative and ablation analyses further show that calibrated edges help retrieval preserve prerequisites, state-changing actions, verification routines, and final completion steps. These results position edge-confidence calibration as an effective route to compact and executable skill retrieval at scale\footnote{Code is available at: https://github.com/ZhiyuanLi218/Caskg }.

Accurate online traffic prediction is essential for intelligent transportation systems, where forecasting must be performed continuously under imperfect sensing conditions. Missing observations and anomalous disturbances make this task challenging, particularly when prediction relies on a single traffic view. This paper proposes a Multi-View Coupled Tensor Decomposition (MVCTD) model for online traffic prediction from imperfect multi-view observations, such as speed, flow, and occupancy. The proposed model uses coupled tensor decomposition to build a structured latent forecasting space, in which shared spatial structures across traffic views and view-specific temporal dynamics are jointly modeled. A group sparse regularization is further introduced to capture correlated abnormal responses induced by real traffic anomalies and thus reduce their influence on forecasts. For streaming deployment, MVCTD performs iterative refinement only on the current latent tensor, while the remaining model variables are updated by lightweight closed-form steps based on summarized historical information, thereby avoiding repeated optimization over the full historical sequence. Experiments on real-world traffic datasets demonstrate that MVCTD achieves accurate forecasts with favorable runtime under severe missingness, confirming its suitability for online traffic prediction.

Reliable access selection in dynamic and heterogeneous indoor-outdoor environments is challenging because instantaneous radio measurements alone cannot capture future QoS degradation caused by mobility, blockage, traffic load, and resource competition. This paper proposes FedQoS, a federated QoS-risk learning framework for predicting the future reliability of candidate access links and supporting access-node selection without centralizing user-level network data. In FedQoS, each access node locally learns from its observed network logs, including radio, traffic, load, and service-context features, while a global QoS-risk predictor is trained through federated aggregation. The learned model estimates the probability of QoS failure for each candidate link, and the controller uses these risk scores to select reliable access nodes under dynamic network conditions. To evaluate the framework, we construct physics-based synthetic indoor-outdoor wireless datasets using the Sionna framework, covering normal traffic, mobility, event-driven congestion, and non-IID client observations. Simulation results show that learning-based access selection substantially reduces the QoS-failure rate compared with signal-based and historical-QoS heuristic methods. FedQoS achieves near-centralized predictive performance and provides clear reliability gains under mild non-IID data while remaining competitive under the more challenging severe non-IID condition. These results demonstrate the potential of federated QoS-risk learning for reliable, data-local access selection in dynamic wireless environments.

Human-robot interaction (HRI) requires robots to interpret human actions early in their execution in order to respond safely, efficiently, and naturally. However, many existing approaches to human action recognition rely either on sparse skeletal representations, which lack fine-grained motion cues, or dense optical flow, which can be computationally expensive for low-latency perception pipelines. In this paper, we propose PoseOFF, a pose-anchored optical flow representation that captures local motion information around human joints to support earlier human intent understanding. By conditioning motion feature extraction on human pose, PoseOFF encodes localised motion dynamics at semantically meaningful body locations, forming a structured motion representation that is explicitly aligned with human kinematics. We evaluate PoseOFF across multiple benchmark datasets and backbone architectures for action anticipation, demonstrating consistent improvements in recognition accuracy, particularly at early observation ratios. Our results show that PoseOFF enables models to achieve comparable or improved performance while observing less of the action sequence, highlighting its effectiveness for early prediction. Importantly, these gains are achieved without requiring full-frame motion processing, making the approach practical for real-time and resource-constrained settings. These findings suggest that pose-centred motion representations such as PoseOFF can enhance the ability of interactive robot systems to infer human actions earlier, supporting more responsive and anticipatory behaviour in human-robot interaction scenarios.

Multimodal Large Language Models (MLLMs) are increasingly deployed in real-world applications, yet how different factors shape their jailbreak vulnerabilities remains poorly understood. Existing benchmarks often couple harmful intent, prompt framing, visual semantics, and instruction carrier within individual jailbreak instances, obscuring the specific sources of observed vulnerabilities. To address this limitation, we introduce MMJailBench, a factorized benchmark that systematically varies and combines these factors under controlled configurations, enabling fine-grained comparison and factor-level attribution. Large-scale evaluations across 16 open-weight and proprietary MLLMs reveal highly heterogeneous and model-dependent vulnerability profiles. Jailbreak vulnerability varies markedly across harm domains, exposing uneven coverage in current multimodal safety alignment. Prompt framing emerges as the dominant source of variation, task-relevant visual semantics systematically increase jailbreak susceptibility with authority-like cues exposing particularly pronounced vulnerabilities, and visually rendered instructions do not consistently increase jailbreak susceptibility relative to direct textual instructions. To further investigate the risks introduced by multimodal context, we conduct diagnostic analyses on a representative open-weight model and identify vulnerability-associated patterns in internal representations and cross-modal interactions. Finally, we develop a modular multimodal jailbreak evaluation suite with full and lightweight configurations, multiple judge options, and multidimensional metrics, enabling reproducible, scalable, and cost-efficient multimodal jailbreak auditing.

Graph retrieval-augmented generation places retrieved subgraphs into the model's context window at query time, paying a recurring token cost and exposing source data on every call. We study an alternative: compiling a knowledge graph offline into a bank of LoRA adapters, one per entity, that serve as a parametric knowledge layer queried by injecting weights rather than text, at zero query-time context cost. On the MetaQA dataset, we find that subgraph-trained adapters encode context-free factual knowledge that generalizes to unseen questions: on single-valued relations the adapter gains $+0.243$ exact-match score over a base model that is nearly blind closed-book ($0.007$), and only the correct adapter recovers this knowledge (an oracle gap of $+0.283$ over the base model). However, the stored knowledge is not recoverable by similarity: given a query with no subgraph, embedding-based and weight-space geometry retrieval both perform at chance, because a semantically neighbouring entity's adapter does not contain the answer - knowledge is stored locally and does not transfer. Weight geometry correlates with subgraph semantics ($ρ= +0.329$) but not with functional retrievability. We quantify the byte and context-token costs against graph retrieval-augmented generation and discuss deployment implications. Our results establish that parametric knowledge graph memory is feasible for storing knowledge, and identify selecting and composing the right adapters by a mechanism other than semantic similarity as the central open problem - motivating a learned, query-conditioned composition mechanism.

Retrieval-Augmented Generation (RAG) has emerged as a powerful architecture for Question Answering (QA) by integrating external information into Large Language Models (LLMs). However, false, inaccurate, and misleading information in news and social media poses a serious challenge to real-world RAG systems, especially in multi-hop QA, where complex multi-step reasoning can be misled by even a single deceptive misinformation segment in the retrieved documents. Existing approaches mainly rely on implicit alignment or explicit regulation, but their limited ability to assess fine-grained information reliability makes them vulnerable to deceptive misinformation that is semantically relevant to the question yet factually incorrect, leading to erroneous answers. To address this limitation, we propose ReliableRAG, which, to the best of our knowledge, is the first reliability-driven framework that mitigates deceptive misinformation in multi-hop QA through fine-grained evaluation of individual triples. ReliableRAG first extracts information segments from source documents and represents them as structured triples. It then quantifies triple reliability by combining query-triple semantic relevance with triple credibility, retaining only the top-$K$ reliable and non-redundant triples. Based on these refined triples, ReliableRAG autoregressively constructs robust reasoning chains to consolidate trustworthy evidence and filter deceptive misinformation, producing accurate answers faithful to reliable information. Experiments on three multi-hop QA datasets show that ReliableRAG outperforms existing methods, substantially improving the factual reliability and robustness of RAG systems under deceptive misinformation injection.

Long Narrative Reasoning is an essential capability for processing and reasoning over complex narratives. While retrieval-augmented generation provides a promising framework, existing methods still face two critical challenges: cognitive islanding and cross-layer evidence disconnection. To address these issues, we propose PonsRAG, a coordinated RAG framework inspired by the biological pons. PonsRAG consists of two key components: Triple-Layer Indexing, which organizes documents into a connected knowledge structure to bridge cognitive islands, and Coordinated Reasoning, which retrieves evidence across distinct layers and integrates cross-layer information into a unified context. We evaluate PonsRAG on four long-context narrative benchmarks, and experimental results show that it outperforms the strongest baseline, achieving a 11.56% relative improvement in average accuracy on multi-choice tasks.

Dataset distillation compresses a large training set into a compact synthetic set while preserving its downstream utility. However, existing methods primarily preserve global image statistics and may overlook the localized evidence essential for fine-grained visual classification (FGVC), such as object parts, subtle textures, and region-specific structures. We formulate fine-grained dataset distillation as budgeted discriminative-evidence preservation and propose Discriminative Evidence Composition (DeCO). DeCO uses attention rollout from a pretrained TransFG teacher to identify informative patches, applies spatial diversification to reduce redundant coverage, and organizes the resulting regions into class-wise evidence banks. Multiple same-class regions are then packed into compact grid-composed images. The teacher is used only for dataset construction, whereas downstream students are trained with standard hard-label supervision without teacher logits. Experiments on CUB-200-2011, FGVC-Aircraft, and Stanford Cars show that DeCO consistently outperforms representative coreset and dataset-distillation baselines under different IPC budgets.

Generative video models now synthesize footage nearly indistinguishable from reality. Their promise as interactive tools hinges on fine-grained control of how objects and the camera move over time, yet each existing approach captures only part of this: camera-parameter methods steer the viewpoint but cannot move objects, 2D-trajectory methods act in the image plane and ignore depth and occlusion, and recent 3D methods add geometry but run only offline at a fixed length. In particular, none combines 3D-consistent control of both camera and objects with real-time, streaming generation. Here we show that camera motion, object trajectories, and depth can be unified into a single 3D point-track representation, from which one model performs joint camera and object control, depth editing, and motion transfer in a single forward pass. To learn this interface at scale, we mine in-the-wild video for 3D motion supervision, yielding OpenVidHD-Motion3D, and encode it with a lightweight Geometric Motion Head that plugs into a pretrained video diffusion model. Because this encoder is temporally separable, we distill the model into a causal streaming student that generates arbitrarily long video in four denoising steps at memory independent of length. This unified design surpasses prior camera-only, 2D, and offline-3D methods in motion-control precision while covering modalities they address only in isolation. 4DStreamCtrl runs at 20 FPS on a single high-end GPU for 480p video and stays temporally coherent over hundreds of frames, enabling, to our knowledge, interactive 4D-controllable streaming generation for the first time. More broadly, grounding generation in explicit 3D geometry with efficient causal inference points toward interactive world models with closed-loop spatiotemporal control, from controllable simulators to real-time visual imagination for embodied agents.

In recent years, distinguishing between AI-generated text and human-written text has remained a challenge. In this paper, we introduce VietAIDetector, an open-source tool designed specifically for detecting Vietnamese AI-generated text. It allows users to interact through a Gradio web interface with inputs ranging from raw Vietnamese text to common text file formats, including scanned documents and exceptionally long texts that exceed the context size of the employed Large Language Models (LLMs). The core component of the tool employs a Zero-Shot approach to detect AI-generated text without requiring domain-specific training data, building upon the previous VietBinoculars and Binoculars research. The tool is built upon a Vietnamese-specific language model and has been evaluated on out-of-domain datasets, demonstrating superior performance compared to existing methods primarily developed for English. Additionally, users can select optimal detection thresholds based on F1 score, accuracy, or TPR@0.05FPR requirements. The results are presented through the web interface, allowing users to easily review and verify suspicious texts or download them as a PDF report. The tool is publicly available at https://github.com/trieuntu/VietAIDetector

Unmanned aerial vehicle (UAV)-enabled integrated sensing and communication (ISAC) is a promising 6G paradigm, but dynamic multi-UAV ISAC control must jointly balance communication quality, sensing reliability, and flight safety under stochastic mobility. Existing optimization methods often require repeated global non-convex solving, while online reinforcement learning (RL) depends on risky trial-and-error flights that may cause sensing loss or collision-risk events. This paper proposes AERIS, an offline policy improvement framework for multi-UAV ISAC. AERIS learns from fixed flight logs under centralized training and decentralized execution, so each UAV acts from local histories while training uses logged global information to assess team-level effects. We further design STAR-CRDT, an offline multi-agent RL algorithm that performs support-aware local action rectification and distills only trusted improvements into the decentralized actor. We prove an offline-support policy improvement guarantee. Experiments show that STAR-CRDT improves the main ISAC objective return by 29.3% over the strongest baseline. It further improves communication sum rate, sensing pass rate, and sensing margin by 3.4%, 4.8%, and 69.1%, while reducing collision-risk events by 54.2%. On unseen real-road maps built from OpenStreetMap data, STAR-CRDT still obtains the best return.

A system that authorizes an action must see enough of it to decide, and a system that attests to its decision must record enough to be audited. Both pressures push raw action parameters -- recipients, payment memos, record identifiers -- into an append-only ledger that cannot delete them. We show the two are separable. We classify each parameter field, not each action class, into three tiers: fields a policy may legitimately match on, which cross raw; fields that are policy-relevant but identifying, which cross only as projections such as an email domain or a templated route shape; and fields with no legitimate policy use, which never leave the workload. The central property is that the ledger's commitment is a canonical digest of the full, unminimized parameters, computed before minimization runs. The commitment is therefore independent of the tier table: reclassifying a field changes what is disclosed without invalidating a historical entry, reopening a hash, or altering what an offline verifier checks. Tier table, policy schema and wire schema are generated from one per-action declaration, so the deciding and recording parties cannot hold different rules. We then address a question the architecture forces: which party should compute each attested fact? We argue it is settled by which party could lie about it undetectably, and derive three answers within one request -- the client computes the parameter digest, being the only party holding the data; it is structurally prevented from naming the definition that governed it, since that would write a false statement into a signed ledger; and it attests which tier table it applied, so divergence is detectable. We give a leakage analysis of each projection, report an incident in which a first-cut projection preserved the identifier it was written to remove, and state the residual trust the design does not eliminate.

Functional data analysis is an important statistical field that treats data as random functions. In practice, the random functions are often not fully observed but instead measured at discrete times. While simpler problems, such as mean and covariance estimation, have been widely studied for discretely observed data, optimal estimation of linear regression for this data type has remained unsolved for over two decades. To tackle this fundamental challenge, we propose a novel approach, referred to as pooling ridge estimation, which combines the advantages of pooling strategy and RKHS-based method by incorporating the unbiased estimation of operators based on discretely observed measurements from all subjects. This unified estimation framework enables us to achieve minimax optimality in prediction risk in arbitrary sampling schemes ranging from sparse to dense designs, for both scalar-on-function and function-on-function regression models. Such methodological and theoretical advances are obtained for the first time and accurately reveal the influence of discrete sampling. For scalar-on-function regression, the phase transition occurs once, separating the convergence behavior into two distinct regimes. Remarkably, for function-on-function regression, up to three phase transitions may occur, determined by the sampling frequencies of the predictor/response functions. Finally, simulation experiments and two real data examples provide empirical support for the proposed methods.

Multimodal regression suffers from the mean-collapse pathology: under squared loss, an unconstrained regressor converges to the conditional mean, which for K > 1 lies away from all modes. We attribute this failure to pairwise contradictions--samples with nearly identical inputs but distant outputs--and propose Difference-Quotient Clustering (DQC), which partitions data to minimize intra-cluster output-vs-input discrepancy. Each sample is assigned to the cluster that minimizes its maximum contradiction ratio; a logits generator and a conditional network are then trained on the resulting labels. Since the generating modality is unknown at test time, we evaluate via minimum squared error (minMSE) against all K true outputs. On synthetic benchmarks (K=5, 10), DQC achieves test minMSE 0.19 (K=5, nx=500), versus 0.09 for an oracle, 1.08 for random labels, and 1.33 for mean collapse. We observe two empirical regularities: larger intra-cluster contradictions require deeper networks, and oracle labels generalize from fewer samples than cluster-derived equivalents. The clustering is a hard, parallelizable O(n^2/2) front-end for coarse conditional assignment, reducing the burden of downstream generative refinement. A second-stage re-clustering on residual errors is outlined as future work.

The functional annotation of genes in non-model organisms remains a significant challenge in computational biology, with 20-70% of sequenced genes lacking characterized functions. Traditional homology-based methods are often costly and strongly dependent on high sequence similarity. This study presents Homo-RAG, a framework for large language model-based gene function prediction that integrates homology-guided multi-hop retrieval with evidence-aware ranking. The framework exploits biological relationships between zebrafish and human orthologs to guide evidence acquisition from ZFIN, UniProt, and PubMed through hybrid dense and lexical retrieval. An Evidence Confidence Score (ECS) integrates semantic relevance, entity matching, orthology information, source reliability, and literature association signals to refine the ranking of retrieved evidence. Extensive evaluation across 150 queries and 7,200 retrieved documents shows that evidence weighting parameter of lambda=0.50 improves NDCG@10 to 0.9879 and MRR to 0.99, while retrieving relevant evidence for 99.33% of queries. Furthermore, 80% of the retrieved documents are query-exclusive, indicating that evidence quality complements rather than replaces retrieval relevance. These findings establish Homo-RAG as a practical and robust framework for reliable, evidence-grounded gene function prediction in understudied organisms. The study addresses important limitations of conventional annotation pipelines while identifying opportunities for future improvements in evidence features and attribution mechanisms.

Visual inspection of welded assemblies remains one of the least automated stages in many industrial production processes, still depending largely on the experience of human operators and thus subject to inter-operator variability; the manufacturing of special-purpose machinery cabins, the setting of this study, is one representative case. This work evaluates the feasibility of automatic weld seam segmentation from RGB and polarimetric imagery, comparing controlled laboratory acquisitions with images captured under real, uncontrolled conditions. Convolutional neural network (CNN) architectures and transformer-based architectures are benchmarked under a unified, threshold-independent protocol, training each CNN with three random seeds to separate genuine effects from seed noise. In controlled RGB conditions, CNN models reach a mean mask mAP50 of up to 0.87, but drop to 0.22-0.48 under uncontrolled acquisition, showing that the acquisition setup is a first-order component of the inspection system. Polarimetric imaging with alignment-preserving geometric augmentation localizes previously unseen welds with a mean mask mAP50 up to 0.93: on par with, rather than ahead of, the best controlled-RGB result, but reaching that accuracy on uncontrolled RGB without requiring acquisition control. The clearest architectural finding concerns viewpoint robustness. In-distribution, transformers and CNNs are broadly comparable; but under a test-time viewpoint shift, the transformer models, and RF-DETR in particular, retain high accuracy while every CNN collapses. The gap holds across three seeds and a resolution-matched control, pointing to architecture rather than training resolution. Within the CNN family, capacity brings no reliable in-distribution gain once seed variance is accounted for: small CNNs suffice for fixed viewpoints, transformers for variable ones.

Alignment auditing of frontier models increasingly relies on LLM auditors to surface undesirable behaviors at scale, but current automated auditors can struggle with coherent investigation and audit realism. In this work, we improve LLM auditors with reinforcement learning. In our best training environment, the policy investigates target models that potentially possess hidden behaviors planted via their system prompt. An LLM judge, which knows whether the target has a hidden behavior, holistically compares the policy's investigation to a reference investigation to determine the reward. With systematic ablations, we find that pairwise rewards yield more robust training compared to pointwise rewards, and that adding targets without planted behaviors helps maintain a low false positive rate. Training improves investigation quality against targets with planted behaviors, the rate of concerning behaviors surfaced in unmodified production models, and audit realism, while false-positive rates stay below 1%. Furthermore, auditing capabilities generalize across scaffolds: performance on AuditBench's adversarially fine-tuned targets substantially improves [Sheshadri et al., 2026].

Despite their strong ability to generate code, large language models often fail to produce secure code, as their outputs frequently contain security vulnerabilities. Secure code generation is inherently challenging because it requires solving a multi-objective problem: functional correctness and security. Existing approaches address this challenge by injecting external security knowledge or by using agentic feedback and iterative refinement. However, guideline retrieval often leaves the generator to translate generic advice into task-specific secure implementations, while shared-dialogue multi-agent feedback can blur role boundaries and suffer from context bloat. We present MACGen, a multi-agent framework that integrates planning, security analysis, code synthesis and refinement to jointly optimize security and functionality. A planner constructs a step-by-step plan to satisfy functional requirements. A security advisor identifies likely CWEs and synthesizes task-specific guidelines, a coder then generates code grounded in these artifacts, and a reviewer issues perspective-separated feedback. Rather than sharing full dialogue histories, each agent receives only structured artifacts from upstream stages, enforcing role specialization and reducing uncontrolled context growth. On CWEval and BaxBench, MACGen improves F&S@1 over direct prompting by 19.61 and 10.57 percentage points (pp) on average, respectively.

We introduce VGA-BenchV2, an extended human-aligned benchmark and optimization framework for jointly evaluating and improving video generation quality and aesthetic value. Built upon VGA-Bench, VGA-BenchV2 preserves the original fine-grained taxonomy with two primary dimensions-Aesthetic and Generation-and 52 sub-dimensions. Guided by this taxonomy, we curate 1,016 diverse prompts and collect over 60,000 videos generated by 12 mainstream video generation models. More importantly, VGA-BenchV2 substantially expands human-labeled supervision by adding 36,000 task-level annotations, including 16,200 for aesthetic quality, 13,200 for aesthetic tagging, and 6,600 for generation quality, corresponding to 13.46x, 11.15x, and 1.55x scale-ups over VGA-Bench, respectively. Leveraging this enlarged annotation corpus, we develop a hybrid evaluator architecture consisting of VAQA-Net for continuous aesthetic scoring and two Qwen-based Large Vision-Language Model evaluators, VTag-Net and VGQA-Net, for aesthetic tagging and generation quality assessment. Extensive experiments demonstrate strong alignment with human judgments across diverse generation models. Beyond evaluation, VGA-BenchV2 further introduces an evaluation-to-optimization pipeline, where the learned aesthetic evaluator serves as a reward model for reinforcement learning-based generator fine-tuning. This closes the loop from benchmark construction and human supervision to automated evaluation and model optimization, enabling video generators to improve not only in realism but also in aesthetic quality and human preference alignment. Resources are available at https://huggingface.co/datasets/BestiVictoryLab/VGA-Bench.

Formal theorem proving enables machine-verifiable evaluation of mathematical reasoning, yet existing benchmarks often emphasize aggregate proof accuracy, concentrate on a narrow range of mathematics, and provide limited evidence of robustness to equivalent reformulations. We introduce MathAdv, a diagnostic benchmark spanning 13 domains across undergraduate- and graduate-level mathematics. Alongside Lean 4 theorem proving, MathAdv provides up to three auxiliary tasks: multiple-choice questions that probe mathematical knowledge, fill-in-the-blank problems that isolate informal reasoning, and expert-crafted transformations that test robustness to problem presentation. Our evaluation of contemporary theorem provers yields four findings: formalization remains a major bottleneck; performance varies substantially across mathematical domains; natural-language guidance helps general-purpose LLMs but can hinder proof-specialized models; and mathematically equivalent reformulations expose substantial robustness limitations. Together, these results show how component-wise evaluation can reveal model capabilities and failure modes that aggregate theorem-proving accuracy obscures. The dataset and evaluation scripts are available at https://github.com/margotyjx/MathAdv.git.

To resiliently and sustainably meet our future energy demand, photovoltaic (PV) modules must be deployed across a broad and diverse range of geographical regions with varying operating conditions. As these conditions strongly affect both performance and optimal system design, a dedicated PV-specific climate classification can be of great use. In this work, we develop a climate classification framework tailored to PV applications using a variety of machine learning (ML) techniques. Building on previous studies, our approach incorporates both energy yield, and for the first time, also the module lifetime with climate dependent degradation. We generate an interpolated dataset containing twelve input features and two target variables (i.e. energy yield and module lifetime). Feature importance analysis shows that annual global horizontal irradiation and ambient temperature are the most influential predictors. The most accurate regression model achieves root mean square errors (RMSE) of 0.007 MWh for energy yield and 1.5 years for lifetime prediction. The calculated feature importance scores are then integrated into a hierarchical clustering framework, resulting in 6 primary climate clusters (Tropical, Desert, Continental, Temperate, Boreal, and Polar) and 15 corresponding subclusters. Our analysis shows that the low temperature continental climate offers the highest discounted lifetime energy yield. These results can support a wide range of applications, including PV module optimization, system siting decisions, and comparative performance studies.

Efficient determination of the effective multiplication factor (keff) is an important computational task in reactor core neutronics analysis. Physics-informed neural networks (PINNs) incorporate neutron diffusion equations and boundary conditions into network training to efficiently determine the neutron flux distribution and keff. To further improve the efficiency of keff calculations using PINNs, a Joint Initialization Physics-Informed Neural Network (JI-PINN) is proposed in this work. In this method, a low-resolution approximate solution to the K-eigenvalue problem is used to construct a joint initial state for the flux network parameters and keff, and both are then jointly optimized under physical constraints. The proposed method was validated on a two-dimensional two-group two-material case, the IAEA 2D benchmark, a two-dimensional two-group four-material case, and a three-dimensional single-group case. For these test cases, the total computational time was reduced by 25.4%, 38.2%, 49.4%, and 28.9%, respectively, while comparable solution accuracy was maintained. The occurrence of anomalous results associated with marked deviations of keff from the reference value was also reduced. The proposed method provides a more efficient and robust initialization strategy for solving neutron diffusion K-eigenvalue problem with PINNs.

Fine-grained segmentation of communication-tower components in UAV imagery is essential for automated inspection, yet task-specific models are hard to develop due to limited instance-level annotations. Zero-shot segmentation models offer a promising alternative, but in cluttered scenes, visually similar background structures interfere with component localization, causing missed instances and false positives. We propose a model-agnostic saliency-depth foreground-conditioning strategy combining appearance-based saliency with monocular relative depth to construct a coarse tower prior and suppress irrelevant content. We integrate this module with Grounded-SAM and SAM 3, yielding SD-Grounded-SAM and SD-SAM 3. SD-Grounded-SAM further applies geometric and depth-aware box refinement before mask generation, while SD-SAM 3 relies on SAM 3's internal setup. On TOW-300, a dataset of 340 communication-tower UAV images, our strategy improves both baselines: SD-SAM 3 achieves the strongest instance-segmentation performance, while SD-Grounded-SAM produces fewer false positives. Ablations confirm complementary gains from saliency, depth, and box refinement, improving robustness in cluttered scenes.

Machine unlearning aims to make a model forget specific data, yet unlearned LLMs often fail to stay unlearned: brief fine-tuning can revive removed knowledge. Existing robustness predictors rely on global weight-space displacement, but distance alone can be misleading when random or destructive updates collapse performance. We argue that relearning robustness depends on update structure: robust unlearning should affect forget-critical weights while sparing retain-critical ones. We introduce the Forget-Retain Alignment Gap (FRAG), a training-free predictor that scores an update's forget-retain alignment without running a relearning attack, and separates selective from dense updates more reliably than global distance. Building on the forget-critical, retain-sparing principle, Forget-Retain Pruning (FRP) improves relearning robustness. Our results suggest that weight selectivity better explains robustness than distance alone. Code is available at https://github.com/Yi1-Chen/FRAG.

Correcting flawed reasoning traces remains a significant challenge for Large Language Models (LLMs), whose autoregressive generation can propagate early mistakes into subsequent reasoning. We introduce DCGC, a Masked Diffusion Model (MDM) framework for global correction that uses an imperfect solution draft from an upstream solver as auxiliary context. DCGC combines task-specific Supervised Fine-Tuning (SFT) with a novel inference-time mechanism called Dynamic Dual-CFG. This mechanism separates problem-only and joint problem-draft branches and scales the draft-conditioned residual using a relative confidence gap. Across math, code, and knowledge reasoning benchmarks, DCGC outperforms standard sampling and simpler CFG variants, with additional results suggesting transfer to different diffusion backbones. In test-time setting where ground-truth failure labels are unavailable, DCGC improves full test set accuracy by correcting low-consensus upstream outputs, highlighting its utility as a verifier-free global correction module for difficult reasoning instances.

Screen rotation is a fundamental interaction in Android applications, but it often introduces non-crashing functional failures (NCFs), such as layout inconsistencies and state loss, which are difficult to detect automatically. A key challenge is the lack of effective test oracles for checking cross-orientation state equivalence between portrait and landscape views. We propose RotDroid, a testing framework for detecting GUI rotation bugs via cross-orientation state equivalence. RotDroid generates and mutates State-Preserving action Sequences (SPS) to construct semantically equivalent GUI states across orientations. To support reliable oracle checking, we build RotBench, a dataset of paired portrait-landscape GUI states, and develop RotVL, a vision-language model fine-tuned for equivalence checking. Experiments on both synthetic and real-world datasets show that RotVL outperforms state-of-the-art models, and RotDroid detects more rotation-induced failures than existing techniques under equal budgets. In large-scale studies on open- and closed-source apps, RotDroid reports 94 previously unknown bugs, with 47 confirmed or fixed by developers, demonstrating its practical effectiveness.

The Stochastic Simulation Algorithm (SSA), widely considered an exact algorithm for stochastic chemical reaction networks, suffers from high computational cost. In this work, we propose a data-driven effective model that operates on a user-defined coarse time step independent of the underlying microscopic reaction-event scale. This is accomplished by directly approximating the finite-time transition kernel of the continuous-time Markov chain induced by SSA, using a generative machine learning model trained on short bursts of SSA simulation data. The trained model constructs a stochastic propagator that recursively generates statistically consistent trajectories at the constant coarse time step, with significantly reduced computational cost. In this paper, we employ conditional normalizing flow as the stochastic propagator. A comprehensive set of numerical examples is presented to demonstrate the accuracy and efficiency of the proposed method.

Beyond-visual-range (BVR) air combat is a challenging reinforcement-learning domain characterized by partial observability, long-horizon decision making, energy management, and limited weapons. We present BVR Sim, an open-source Gymnasium-style environment designed for heterogeneous air-combat reinforcement learning. BVR Sim supports multiple JSBSim aircraft models, including the F-15, F-16, F/A-18, and F-22, with configurable weapons, sensors, controllers, and opponents. A unified tactical action interface specifies desired heading, altitude, speed, and weapon release above aircraft-specific inner-loop controllers, enabling policies to operate across heterogeneous platforms. The environment provides interchangeable Python and accelerated C++ backends, entity-oriented observations, compositional rewards, scripted opponents, replay and visualization, and adapters for multi-agent learning frameworks. At a 0.4-s decision interval, the C++ backend achieves 104 simulated seconds per wall-clock second in 1-vs-1 and remains practical through 10-vs-10 scenarios. A policy trained only on the F-16 transfers without retraining to four unseen aircraft, reaching a 45.5% mean win rate with aircraft-specific controller adaptation. MAPPO and HAPPO experiments further verify end-to-end compatibility with standard multi-agent reinforcement-learning pipelines.

Evaluation is shifting from static QA toward agentic settings where models act through external tools. We identify a critical yet underexplored capability within this space - dexterous visual tool use: fine-grained, closed-loop parameterized visual action in which models infer tool parameters from visual evidence, and those parameters directly govern the final result. Existing benchmarks cover web navigation, GUI operation, and software engineering, but rarely target this coupling between visual evidence and execution precision. We propose EASEL, a benchmark evaluating a controlled instance of dexterous visual tool use that adopts reference-guided visual reconstruction as its primary proxy task: the agent incrementally paints a canvas to match a reference image. EASEL additionally includes semantic tasks spanning region annotation, handwriting, and path planning. We further provide EASEL-Data, a 440k-sample two-stage curriculum dataset for trajectory supervision, and EASEL-9B to investigate its effect on this capability. Evaluation of 25 models reveals that current multimodal agents systematically struggle on EASEL. Reconstruction similarity bottlenecks at low levels (0.40-0.54), while trajectory diagnostics expose severe closed-loop instability - models typically saturate early or degrade post-peak. Semantic tasks reveal sharp capability boundaries in precision annotation and path planning. EASEL-9B, trained on EASEL-Data, surpasses the base model by a relative 6.3%, ranking third among all evaluated models.

Token communications realize the semantic communication principle at the granularity of transformer tokens, providing a promising direction for client--server collaborative inference in resource-constrained edge systems. However, directly transmitting token embeddings presents two practical challenges: substantial communication cost and limited interoperability across model-specific token embedding spaces. To address these challenges, we propose a \emph{token-oriented} semantic communication framework. In this framework, token-level task relevance determines which compressed image latents are transmitted, enabling token-granular transmission without directly transmitting token embeddings. The framework is modular, coordinating three pretrained components---a lightweight client-side vision transformer (ViT), a learned image compression (LIC) model, and a large server-side ViT---without end-to-end training. The key enabler is the one-to-one spatial alignment between ViT patch tokens and the LIC latent vectors, which allows token-level task relevance to directly determine which latent vectors are transmitted. Building on this alignment, token-aligned LIC selectively transmits task-relevant latents, layer-selective attention rollout estimates token relevance from a selected range of attention layers in a single forward pass, and surrogate token substitution adapts the frozen server model by optimizing a single learnable token. Experiments on ImageNet show that the proposed framework achieves a more favorable rate--accuracy trade-off than recent semantic communication schemes, hand-crafted codecs, and task-agnostic LIC models.

Retry mechanisms are a standard component of resilient distributed systems, but their collective behavior, when every tier in a call path retries concurrently, is less well understood than the per-client guidance that produced them. This paper introduces the retry amplification factor (RAF), a metric quantifying the additional request volume that retry policies generate during partial failures. In a study of 200 open-source Python microservice projects, explicit retry logic is detected in 11.5%, and an audit of our own false negatives places true prevalence near 41%. Among the projects detected, 60.9% contain at least one configuration without backoff, and after manual verification exactly one of 113 production configurations randomizes its delay. We then evaluate these policies in simulation (n = 100 trials per strategy). Under correlated failures, a naive standard retry policy reduces the success rate from 55.4% to 41.5% relative to performing no retries at all. We catalog five recurring anti-patterns, propose Adaptive Retry Budgeting (ARB), and show that budget-constrained retries maintain success rates close to the no-retry baseline while still recovering from transient faults. These results indicate that retry behavior should be designed as a system-level property rather than configured locally at each call site.

Neural radiance fields (NeRFs), 3D Gaussian Splatting (3DGS), and related novel-view synthesis methods are commonly evaluated under capture and reconstruction conditions cleaner than those encountered by robots, drones, and autonomous systems. Benchmarks often rely on reconstruction-friendly trajectories, optimized camera poses and intrinsics, and held-out views sampled from trajectories represented during training. These assumptions can obscure performance with measured poses, reusable camera calibration, and structurally different camera paths. We introduce PIVOT (Pose, Intrinsics and Viewpoint Oriented Testbed), a multi-trajectory dataset, processing pipeline, and evaluation framework for independently studying these factors. PIVOT captures each scene using diverse camera trajectories and retains, where available, both sensor-derived measured poses and COLMAP-optimized poses, together with calibrated and optimized camera intrinsics. It defines three benchmark families: (1) seen versus unseen trajectory novel-view generalization, (2) measured versus optimized pose sensitivity, and (3) calibrated versus optimized intrinsics sensitivity. We also introduce a directed pose-space Chamfer distance to quantify how well training poses cover an evaluation trajectory. PIVOT v1 contains five real-world scenes captured with a DJI Mini 4 Pro and provides an open processing and Nerfstudio-based evaluation toolchain. Benchmark results show a consistent quality gap between held-out views on represented trajectories and unseen trajectories, as well as substantial sensitivity to pose source and camera intrinsics.

Agentic coding workflows are now widely deployed in real-world systems. With long-horizon reasoning and tool use, token usage has become an important consideration for both cost and efficiency. Two engineers using AI will solve the same problem differently. How the specification of a task shapes an agent's token spend, and whether that spend can be predicted in advance, are open questions. Here, we study the effects of different task specifications on agentic token spend with the Kimi K3 model at three thinking efforts. Across $2,700$ runs, we show that reducing a full task specification to a bare user story raises token spend by $29.7\%$, while run-to-run variance remains unaffected by any prompt changes. We show that prompt-sensitivity is task-dependent, running from $13\%$ to $115\%$. We fit a simple predictor that can price a full distribution of task specifications and thinking effort configurations from a single cheap probe on an unseen task within $36\%$, improving over prior work in predicting token spend. Our work provides initial results quantifying the effects of task specification on agentic token spend and introduces a method that can be used to systematically evaluate the cost of AI coding workflows.

Multimodal Large Language Models (MLLMs) have demonstrated strong abilities in solving diverse visual and textual reasoning tasks. However, their development in the physics domain is significantly hindered by the lack of a comprehensive benchmark. To fill this gap, we introduce OmniPhys, a large-scale benchmark for multimodal physics understanding and reasoning, covering middle school through university-level problems from Chinese Educational Corpora. OmniPhys consists of 15,246 questions and 19,850 images, accompanied by detailed annotations that support fine-grained analysis of reasoning processes and knowledge usage. Beyond conventional evaluation, OmniPhys is a benchmark that systematically evaluates multimodal outputs in the physics domain, including models' ability to generate structured physics diagrams, which constitute a fundamental component of authentic physics problem solving. Extensive evaluations reveal critical gaps in the capabilities of current MLLMs, especially in complex reasoning and visual generation. To address this, we release OmniPhys to serve as a foundational resource for advancing multimodal intelligence in physics and scientific domains. Codes and data are available at https://github.com/ECNU-RAIL/OmniPhys-EMNLP2026.

Refusal training protects AI models from jailbreaks by training models to decline unsafe queries, reducing the risk of misuse. Recent work finds that refusal behavior in aligned language models can be mediated by a single activation direction or a low-dimensional refusal subspace shared across harmful prompts: ablating those directions suppresses refusals while largely preserves other model capabilities. Yet it remains unclear why safety-critical features in a wide range of models emerge and concentrated, low-dimensional structure. In a case study of OLMo-2-0425-1B-Instruct we find that the refusal geometry reflects refusal training: activation updates resulting from refusal-completion first-token losses explain the resulting refusal direction and refusal subspace. We study refusal directions through the training dynamics across refusal datasets and reveal that their brittleness is associated with repetitive refusal starts, which in turn is linked to concentration of gradients and refusal features in a low-dimensional subspace. Across frozen-model analyses and controlled synthetic fine-tuning, we find evidence of a hardening lever: diverse refusal starts can raise stable ranks of gradients and activation changes, making refusals harder to remove with a vector ablation attack.

We describe a systematic approach for spawning and aggregating multi-class cryo-EM reconstruction jobs. This approach formalizes standard ad hoc strategies of iterative classification and filtering typically used by practitioners to sort impure, heterogeneous samples. To our knowledge, this is the first method that can successfully perform ab initio reconstruction on datasets containing dozens of distinct species. We obtain 97% accuracy on ab initio reconstruction of a 45-class subset of Tomotwin-100, 75% accuracy on the full Tomotwin-100 dataset, and demonstrate recovery of ribosomal assembly states from an unfiltered experimental cryo-EM dataset. Our approach's capability scales with compute and lays the foundation for automated cryo-EM workflows in modern experimental settings.

Blind and low-vision (BLV) users are increasingly engaging with large language model (LLM) interfaces to access documents, but it is unclear how such systems support or hinder their ability to build interconnected knowledge. To examine this gap, we compared a Question-Answer Interface (QAI) that supports open-ended conversational inquiry, with a Document Interface (DI) based mostly on traditional structured text document navigation. We recruited 16 BLV screen reader users where they used both interfaces to explore two fictional worlds. Data from interaction logs, concept maps, decision-based tasks, and semi-structured interviews provide comparative insights into how interface design supports knowledge construction. Findings show that participants visited more distinct documents with the DI and formed larger and more correct mental models with the DI than with the QAI. They were also more able to apply knowledge they had gained. Simultaneously, many still preferred the QAI and often estimated that they had explored more, formed better mental models and applied their models better when acquiring the information with the QAI, despite this not being the case. Our analysis suggests possible interface design reasons for these differences and highlights some of the risks introduced by using question-answer interfaces to access information spaces.

Chain-of-thought prompting can expose and amplify demographic stereotypes within an LLM's intermediate reasoning and create a failure mode that final-answer debiasing alone cannot address. Mitigating such bias during generation presents a fundamental timing problem: intervening too late allows biased reasoning to propagate, while unnecessarily intervening can disrupt otherwise correct reasoning. Existing approaches largely avoid this decision by either evaluating completed reasoning chains post hoc or intervening at predetermined steps, leaving open when a developing reasoning trajectory provides sufficient evidence to warrant correction. We formulate this decision as an online change-point detection problem. A per-step bias signal updates a CUSUM statistic and a targeted correction is injected only when accumulated evidence crosses a detector-specific threshold calibrated on held-out data. We instantiate the framework with a white-box signal derived from next-token probabilities and a black-box signal obtained from an LLM judge, enabling deployment with both open-weight and hosted models. On gpt-4o-mini adaptive black-box triggering recovers most of the disambiguated-context accuracy lost under fixed-interval intervention while requiring substantially fewer interventions. That result holds even with an independent judge. Across six open-weight models, the white-box signal improves ambiguous-item accuracy on all six but reduces disambiguated-item accuracy on five because it cannot distinguish unsupported stereotype reliance from correct, stereotype-congruent evidence.

Generative vision-language models (VLMs) are increasingly used in human-centered settings, yet they can produce demographically biased outputs even when images differ only in controlled attributes such as perceived race or gender. However, existing inference-time debiasers were largely designed for static embeddings or CLIP-like models rather than generative VLMs. We propose GGSS---Geodesic-Gated Spherical Steering---a norm-preserving intervention that discovers a counterfactual bias subspace on the unit hypersphere, steers visual tokens along geodesic arcs, and uses an adaptive gate to focus correction on tokens that carry stronger demographic signal. We evaluate four generative VLMs against ten adapted inference-time debiasing baselines and prompt-based mitigation under a single operating-point protocol across categorical, pairwise, and occupation-gender bias tests, while also measuring general visual-language capability. GGSS achieves the lowest average bias on all four models, significant on three of four backbones under paired permutation tests, while preserving MMStar accuracy within +/- 0.6 p.p. of the unsteered baseline. Code is available at https://github.com/dukesun99/GGSS.

Sequential recommendation models, while powerful, have limited flexibility in responding to immediate user requests, making it difficult to adapt their recommendations to the user's timely interests. Unfortunately, existing user request adaptation methods often incur high computational overhead due to either 1) retraining the entire backbone network or 2) leveraging the inference ability of large language models (a.k.a. prompt engineering), limiting their applicability in large-scale recommendation services. This paper presents Control via Request-Aware Masking for Editing Recommenders (CRAMER), a framework that takes users' natural-language requests to immediately change sequential recommendation models' behavior. Specifically, inspired by the model control theory, CRAMER treats user requests as control signals to modulate frozen backbone parameters through masking, achieving instant adaptation to diverse requests while avoiding costly retraining. Experiments on multiple large-scale benchmark datasets show that CRAMER outperforms four state-of-the-art request-aware baselines across multiple recommendation metrics while achieving minimal overhead. Moreover, the proposed framework exhibits enhanced controllability and cross-domain adaptability, establishing a new paradigm for request-aware sequential recommendation.

Noisy node classification problem is a fundamental yet challenging task for real-world graph-related web services, where node labels are often corrupted or unreliable due to weak supervision or automatic annotation. However, existing methods typically train models based on one-hot labels, which not only makes models susceptible to overfitting on noisy labels, but also leads to error accumulation after pseudo-label-guided enhancement. In this paper, we propose a novel Partial label-based Self-training framework (PaSta for short) that leverages partial label learning technique to overcome the limitations of existing methods. Specifically, PaSta first trains multiple annotators to comprehensively capture the class distribution of nodes and aggregates their predictions to construct high-quality partial labels. Subsequently, we design a partial label-based classification model with two well-crafted loss functions to guide the model learning at both label and representation spaces. To further enhance the robustness against noisy labels, we introduce a self-training strategy where the labels refined by partial label learning are then used to further optimize the annotators in a closed-loop iterative manner. Extensive experiments on five datasets demonstrate that, compared with existing state-of-the-art methods, PaSta achieves an average improvement of 1.1% in classification performance under various noise settings.

Conversational derailment forecasting aims to predict when online discussions will escalate into hostility, enabling proactive moderation. Existing approaches often struggle in low-data settings and to generalize across domains. This poses a challenge for new platforms and smaller communities where annotated data is limited. We propose modeling pragmatic representations of conversations to reduce lexical noise and improve generalizability. Specifically, speech act information is used as an auxiliary learning signal alongside textual semantics. Experimental results show improved performance across three datasets, particularly in low-data and cross-domain settings.

Structured outputs such as JSON and tables are central to modern LLM-based systems, yet generation failures are evaluated monolithically, conflating two distinct error modes: placement errors (correct values at wrong positions) and value errors (wrong values at intended positions). We introduce Structure-Content Decomposition (SCD), a framework that independently measures structural fidelity and content accuracy. Applying SCD to nested JSON and table tasks across six models (7B to frontier), we uncover a consistent phenomenon: structural fidelity degrades earlier and more sharply than content accuracy as complexity increases. At the highest complexity, even DeepSeek-V4-Flash (with reasoning) misplaces 35% of recalled values, while Qwen2.5-7B misplaces 74%. Controlled ablations suggest that this pattern is associated with reliance on semantic shortcuts rather than topological understanding of output structure. Based on these findings, we propose SA-RLVR, converting SCD metrics into verifiable rewards for reinforcement learning via GRPO. SA-RLVR successfully optimizes structural addressing across distinct topologies: it lifts JSON Value Placement Accuracy (VPA) from 26% to 63% while generalizing to held-out schemas; moreover, it consistently drives VPA improvements in the table domain, demonstrating that structure-aware rewards can directly enhance multi-domain structural positioning.

Model merging provides an efficient way to construct multi-task generalist models without additional training, but its performance often degrades under severe task interference. Task interference in model merging primarily stems from \textit{superposition}, where task-specific features become entangled within the parameter space. This entanglement renders conventional decomposition methods insufficient for effectively isolating useful task directions from interfering components. In this paper, we propose a sparse-representation-based merging framework that uses Sparse Autoencoders (SAEs) to project task vectors into a high-dimensional sparse feature space, enabling feature-level disentanglement before fusion. To reduce computational overhead, we further introduce a lightweight Group-Ranked Zeroth-Order Optimizer (GR-ZOO) to identify task-critical layers for selective merging. Experiments on both Qwen2.5-1.5B and Qwen2.5-7B demonstrate that our method consistently outperforms representative baselines, including Task Arithmetic, TIES-Merge, DARE, Fisher-Merge,and several recent training-free merging methods, across mathematical reasoning, code generation, instruction following, and general knowledge tasks. In a highly conflicting four-task setting on Qwen2.5-1.5B, our method further achieves a 2.78\% improvement over the strongest baseline.

Vision-language models (VLMs) have emerged as a powerful source of supervision for reinforcement learning, enabling agents to leverage rich semantic knowledge during training. Inspired by the success of preference-based reward learning (PbRL) in reinforcement learning from human feedback (RLHF), vision-language model generated image-based preferences provide an effective source for learning reward functions. This can be done by visually comparing two outcomes through the Bradley-Terry (BT) model. However, this pairwise formulation utilizes only two observations at a time, despite VLMs being capable of ranking multiple candidates. The Plackett-Luce (PL) formulation can shape a reward model with listwise rankings as opposed to pairwise preferences, allowing for a more suited use of a VLM based ranking. In this work, to our knowledge, we introduce the first framework that combines VLM-generated preferences with the Plackett-Luce model for reward learning. We evaluate our approach on Meta-World manipulation tasks and show that Plackett-Luce (PL) reward models can train robotic policies from VLM-generated rankings as effectively as pairwise Bradley-Terry, $K$-wise Bradley-Terry, and RL-VLM-F baselines. Across all environments, at least one PL ranking size ($K \in \{3,4,5\}$) consistently performs with or outperforms other methods in mean success rate. Unlike pairwise methods, which are restricted to $K=2$, PL supports different ranking sizes and can therefore be adapted to the environment and desired feedback format. Our best PL configuration achieves an 86% mean final success rate and matches the Oracle baseline on Drawer Open. Overall, these results demonstrate that listwise VLM preference supervision is a competitive and flexible approach to reward learning for reinforcement learning.

Public cryptocurrency archives may appear usable when files exist, although factor research requires observations available and executable at each decision time. We audit public Binance BTCUSDT USD-M perpetual-futures data using event, publication, and availability times and separate proposal from deterministic auditing, evaluation, and holdout access. An initial gapless five-minute requirement for trade, mark, index, and open interest failed: the longest unrepaired intersection was 304.5729166666667 days. A disclosed revision made trade, mark, index, and realized funding the core streams and made open interest optional because its publication time was unverified. The revised mask retained 727 complete UTC days and supported a 436/145/146-day train, validation, and historical-holdout split. On 80 frozen known-rule templates, the auditor detected 40/40 violations and rejected 0/40 legal templates. Across ten null-signal paths, full auditing reduced mean false passes from 0.2910 to 0.0625. Under matched valid-candidate budgets, the audited adaptive agent tied random search and did not establish superiority. In the one-time historical holdout, all evaluated runs had positive IC but negative net Sharpe under primary costs. We therefore report a scoped negative result rather than a profitability or agent-superiority claim.

The Jacobian lens (J-lens) has been proposed as a way to read verbalizable representations from language models. However, its principle and meaning lack a detailed and theoretical discussion. We provide a mathematical view of this interpretation and of its assumed causal structure. Besides treating the J-lens as a heuristic probe, we further regard it as a first-order causal transfer operator from intermediate activations to expected future readouts. We study the Jacobian matrix as the optimal local linear approximation of the downstream mapping, analyze its global approximation behavior and bias, and identify its mathematical meaning as an expectation over anticipated future readouts. Further analysis of the Jacobian energy distribution reveals that its causal geometry is highly sparse. The energy decays with depth, concentrates in an extremely small proportion, and decomposes into diagonal pathways and specific critical positions. This decomposition further resolves the expectation of the J-lens over future outputs into short-horizon and sparse concept predictions, providing a more intuitive attribution and explanation for the ability of the J-lens to visualize concepts during the thinking process. Based on the theory, we propose a simple but effective improvement strategy and decoupling method for the J-lens, which significantly enhances the ability of the J-lens to read out correct intermediate concepts.

Perceived risk in driving evolves over time and may be supported by specific scene entities, yet supervision is typically limited to coarse video-level judgments. Learning \emph{when} supporting evidence emerges and \emph{which entities} support a risk predictor would ordinarily require costly temporal- and entity-level annotations. We introduce \textbf{CoRE}, a weakly supervised coarse-to-fine framework that learns fine-grained prediction support from coarse video supervision. CoRE first trains a video-level predictor and then freezes it. Structured interventions over candidate temporal regions or entity tracks measure how each candidate changes the coarse prediction, producing graded prediction-effect targets. These targets are distilled into a student that directly predicts temporal and entity support from the original video, without requiring interventions at inference. We evaluate this learning principle across three complementary settings: RISEE tests perceived-risk support from subjective clip-level judgments without temporal or entity-level risk annotations; DoTA provides independent temporal event annotations for evaluating weakly supervised traffic-anomaly localization; and UCF-Crime tests whether the same coarse-to-fine mechanism extends to a standard non-driving anomaly-detection benchmark. Across these settings, CoRE learns informative fine-grained support from coarse supervision, with strong temporal localization on DoTA and competitive performance on UCF-Crime. These results show that coarse video predictions can provide useful supervision for recovering the fine-grained evidence supporting them, without requiring corresponding fine-grained labels.

Chinese query correction (CQC) is important for search and query recommendation on content platforms, but supervised methods rely on large annotated correction pairs that are costly to maintain as query vocabularies evolve. Unsupervised correction with language models is attractive, yet in the short-query setting, unconstrained generation often over-corrects ambiguous inputs toward high-frequency phrases, causing intent drift. We propose \textsc{GUIDE}, a generative unsupervised framework for CQC based on a confuse-then-clarify paradigm. \textsc{GUIDE} encodes phonetically or visually confusable characters with shared-IDs and reconstructs the original query with an encoder--decoder architecture, which constrains correction to plausible confusion neighborhoods while learning from unlabeled query streams. A time-decayed, query-frequency-weighted objective further supports adaptation to rapidly changing query vocabularies. Experiments on \textit{QSpell 250K} and a large-scale real-world dataset (\textit{KwaiSearch}) show that \textsc{GUIDE} consistently outperforms strong baselines, while online A/B testing further confirms gains in correction quality and downstream engagement.

Large language models (LLMs) can fluently verbalize statistical evidence, yet statistical reports can still drift numerical values, invert effect directions, or restate thresholded contrasts as categorical effects. We frame these failures as a control problem: the evidence-bearing content of a scientific report should be fixed by structured statistical results rather than sampled during prose generation. We therefore use cross-run reproducibility to stress-test whether report-visible numbers and claims are bound before prose generation. Existing controls operate at the text or slot level; a deterministic hybrid template reproduces only 61.1% of report-visible numerical content across seeds because the LLM still selects which findings and numbers the template renders. We propose claim-locked reporting, a provenance-before-prose protocol that fixes the evidence source, numbers, direction, and allowed language strength of each reportable claim before the LLM writes only connective prose. Across fMRI functional-connectivity reporting and randomized controlled trial reporting on Evidence Inference 2.0, claim-locked reporting improves reproducibility over the hybrid template by 37.4 and 20.5 points, respectively. Blinded human audits support the observed direction-preservation and governance trends. In an fMRI cost analysis with DeepSeek, claim-locked reporting also yields the lowest observed token use and median generation latency.

Memory-augmented agents maintain compact user profiles throughout extended conversations, enabling personalized and consistent responses without the need to process the entire dialogue history. The quality of these user profiles relies on the underlying memory management strategy: at each step, the agent must determine what to retain, compress, or discard. However, existing methods typically employ a static, one-size-fits-all strategy established before training. In practice, the optimal memory decision is inherently user-specific and dynamically evolves alongside policy optimization. To address this, we propose \textbf{HiPS} (\textbf{Hi}erarchical \textbf{P}ersonalized \textbf{S}trategy), a framework that decouples memory management into a globally shared foundation and a user-specific adaptive tier. Specifically, HiPS employs \textbf{Universal Strategy} to extract shared principles from cross-persona trajectories, alongside \textbf{Persona Delta Distillation} to generate tailored rules for users whose behaviors diverge from general patterns. \textbf{Cross-Level Rule Flow} dynamically calibrates their boundary by promoting broadly validated personal rules and demoting contradicted global ones. The architecture establishes a co-evolution loop where a mechanism guarantees that all strategy refinements are anchored to task outcomes. Extensive experiments demonstrate consistent improvements over memory-augmented baselines.

Physics-Informed Neural Networks (PINNs) frequently fail on stiff or advection-dominated PDEs, and two recent accounts offer competing remedies: switching from FP32 to FP64 to repair an L-BFGS stopping artifact, or replacing the MLP with a state-space-model (SSM) backbone plus sub-sequence alignment to counter architectural simplicity bias. We test both under matched, seed-paired controls in a pre-registered 144-run study spanning convection, reaction, and wave, plus an independent 85-run convection/wave study; success is relative $\ell_2$ error below $0.05$. The two remedies act on disjoint regime-and-seed slices: neither substitutes for the other. On hard convection ($β{=}50$), alignment recovers 2/5 seeds in FP32 and 3/5 in FP64, where the unaligned SSM succeeds on 0/5 seeds at either precision and the vanilla MLP moves only from 0/5 to 1/5 across the precision switch---the recoveries trace to the alignment objective, not the backbone. On reaction the backbone alone already succeeds on 3/5--4/5 seeds, so each remedy covers a regime the other does not. Responses are also seed-specific: the same precision switch flips individual seeds in opposite directions and, on wave, lowers median error with no statistically significant success gain. Tightening the inner L-BFGS tolerance in an independent repeated-step runner likewise lowers median error at a large runtime cost, with success counts unchanged. Precision, stopping, backbone, and alignment must therefore be evaluated jointly and reported per seed.

In transductive classification, an adversary fixes a labeled population, one label is hidden uniformly, and the learner sees all remaining labels. For binary classes, agnostic transductive and PAC learning have the same minimax rate. Whether this extends to multiclass learning was open, especially for unbounded label spaces where uniform convergence can fail. We resolve the question up to logarithmic factors. For every multiclass class $\mathcal H$ with DS dimension $d_{DS}$ and Natarajan dimension $d_{\mathrm N}$, the optimal agnostic transductive excess error satisfies $\widetildeΘ\left(\frac{d_{DS}}{n}+\sqrt{\frac{d_{\mathrm N}}{n}}\right).$ The result holds for arbitrary label spaces. The two terms are both necessary. A DS pseudo-cube gives the realizable $d_{DS}/n$ obstruction, while a Natarajan cube with repeated points and fair labels gives the agnostic $\sqrt{d_{\mathrm N}/n}$ obstruction. The upper bound uses a random-reservation principle. The learner deliberately ignores a constant fraction of the visible labels, which makes the true test point uniform in a large unseen block. We combine realizable compression, a label-space reduction, and inside-menu agnostic compression across this finite-population split. A new without-replacement multiplicative-weights lemma preserves the fast $d_{DS}/n$ term. Consequently, agnostic multiclass PAC and transductive learning obey the same two-dimension law up to logarithmic factors.

Deploying large language models for professional financial review requires more than measuring general financial competence: models must perform the specific review operation required by a workflow and determine whether available evidence is sufficient for a defensible decision. Existing financial benchmarks cover knowledge, reasoning, compliance, and professional tasks, but their evaluation units are often organized around datasets or task formulations rather than the decisions that deployed systems support. We introduce FinRiskAtlas, a Chinese-language benchmark that evaluates financial LLMs along two complementary dimensions: operation execution under fixed evidence states and evidence-state control under evolving review conditions. The static benchmark contains 9,742 instances across 53 task families, including 42 Domain Knowledge families and eleven downstream review operations defined by explicit evaluation contracts. FinRisk-Ask extends this framework through offline replay of 680 pre-action states from 104 de-identified professional trajectories, withholding future evidence during inference and using it only to construct expert-verified evidence targets. Across 33 model configurations, operation-level evaluation yields non-redundant rankings (mean pairwise Spearman correlation 0.42 across downstream operations), and knowledge-based shortlisting can incur up to 18.01 points of regret on individual operations. FinRisk-Ask further shows that entering the Ask branch more frequently does not necessarily improve request targeting or end-to-end evidence acquisition. These results show that broad financial capability scores do not fully capture where models are reliable in professional workflows, motivating evaluation units aligned with the decisions and evidence states that deployed systems must support.

Software agents connect probabilistic model output to operations that change repositories, processes, networks, and graphical applications. We present Metis, a multi-provider runtime that converts provider streams into typed events before admitted calls reach external effects. Its execution path makes permission decisions, interference classes, terminal results, and lifecycle transitions explicit and inspectable. We evaluate these mechanisms on frozen source artifacts. Across 30 matched real-I/O pairs, four-class mediation reduced median elapsed time from 25.958 ms under forced serialization to 14.146 ms. The mean paired difference was -12.295 ms (95% bootstrap interval [-12.968, -11.694]), with mediation faster in all pairs. A ten-case fault matrix exposed duplicate-identifier and rollback limits. In a child-boundary ablation, the full gate-plus-registry condition blocked the declared unauthorized effect and hid all five escape tools. Removing both protections reversed both observations. A decision-only permission oracle matched all ten declared cases across five invocation routes. Five model conditions also completed a fixed Read-marker protocol in 3/3 trials each. These results support bounded claims about dispatch, permission routing, child authority, and provider-valid trace closure. They do not establish model competence, semantic safety, rollback, or superiority over another runtime.

The move of LLM inference to edge AI accelerators introduces new physical vulnerabilities. During execution, model parameters and intermediate inference states are repeatedly loaded into and processed on the chip, making them suscep- tible to physical side-channel attacks. In this work, by deploying laser voltage imaging, we show that one can extract LLM assets during inference, namely embeddings, attention, and quantized MLP weights, activations, and other inference states, from localized memories and compute subcircuits. To validate our claims, we perform an attack on an FPGA-based LLM accelerator. Since such accelerators reuse the same buffers and compute subcircuits across addresses, tiles, modules, and layers, reading asset values comes down to probing different memories during inference. We demonstrate full recovery of the targeted values; however, we also establish a methodology to recover asset values even if some weights or bits remain unread. We further derive lower bounds that relate imaging effort to asset dimensions and show that even direct recovery scales linearly with the size of the targeted asset

Order-dependent activation statistics are often interpreted as evidence of interaction, but that interpretation can be confounded by where interventions enter the network. We introduce a no-fit site-asymmetry audit. For a twice-differentiable readout, the open-path order-swap decomposes into a canonical additive response measured by single interventions and an antisymmetrized second difference free of first-order and pure self-curvature terms to second order. Across six open-weight language-model families, the single-intervention baseline explains 84.3-97.7 percent of the bracket norm (mean 93.7 percent), while the no-interaction self-curvature term is 1.8-5.2 times larger than the corrected residual in the two families with the plus/minus injection split. The corrected residual clears a generic-interaction null in three of six families under a confound-free prompt split and two of six after configuration robustness. A known-positive surrogate recovers planted mixed interaction, while a matched site-separation test changes the baseline share and a random architecture reproduces the first-order regime. The same estimator transfers to released non-language references: trained residual fractions fall below a fixed Gaussian-direction null in 11/12 contrasts (5/6 ViT-B/16, 6/6 ResNet-50), a portability check rather than pooled evidence. The contribution is a reusable measurement criterion: run the single-intervention baseline before reading an order-swap vector as interaction or geometric structure; if it explains the vector, form the second difference instead. All claims are scoped to activation-space interventions at distinct sites; we do not claim that representation geometry is globally Abelian.

Diffusion Language Models (DLMs) have recently become increasingly competitive with autoregressive (AR) models, and even outperform them on certain tasks. Unlike AR models, DLMs produce output through iterative denoising without a left-to-right order. To further improve the performance of DLMs, we introduce PDC (\emph{Prefix-Denoising Consistency}), a test-time self-verification method for DLMs. PDC exploits a distinctive test-time signal in DLMs under prefix conditioned regeneration, correct trajectories are more stable and reproducible than incorrect ones. Concretely, given an initially generated sample, PDC splits the sentence at an intermediate position and regenerates the remaining tokens conditioned on the fixed prefix. Across mathematical reasoning and commonsense reasoning benchmarks, PDC consistently improves upon the initial sample, outperforms independent generations under a computational constrained comparison, and is robust to different unmasking strategies and parameter settings. These results highlight prefix-conditioned regeneration as an effective DLM-specific primitive for test-time verification.

Multinomial choice models allow flexible substitution patterns but become computationally demanding with many alternatives or observations. With a fixed per-observation simulation budget, simulated maximum likelihood introduces simulation bias, while each optimization step requires a full-sample likelihood evaluation. We propose Stochastic Approximation with Unbiased Simulated Scores (SAUSS), an averaged stochastic approximation based on conditionally unbiased mini-batch score estimates. Each iteration uses a fixed mini-batch regardless of sample size. For multinomial probit, accept-reject sampling provides exact conditional draws and unbiased score estimates for any fixed number of accepted draws. Under local conditions, asymptotic theory for the averaged estimator and the partial-sum process of the SAUSS iterates incorporates mini-batch and simulation variability and supports random-scaling and plug-in inference. In simulations and an application, SAUSS gives comparable results in less than 1% of the computation time of simulated maximum likelihood. SAUSS extends to limited dependent variable models with conditional-expectation score representations and exact conditional sampling.

Guided depth super-resolution (GDSR) typically extracts RGB guidance features through convolutional hierarchies, inheriting their fine-to-coarse bias. Thus, low-level spatial cues surface in early layers, leaving the deeper layers to suppress those that do not correspond to true depth boundaries, which risks artifacts and blurred edges. The same fine-to-coarse bias persists in semantics-based methods that consume low-level tokens early and global tokens late. We present WAVE, which introduces a multi-level discrete wavelet transform (ML-DWT) as an explicit and interpretable feature-control mechanism, enabling a coarse-to-fine reconstruction by consuming sub-bands and semantic tokens in reverse of their generation order. WAVE further exploits these sub-bands to treat high- and low-frequency content separately, filtering at its source the misleading RGB color and texture cues that often lead to blurred boundaries and artifacts, offering an intuitive alternative to the suppression learned implicitly by an opaque network. WAVE separates structure and detail reconstruction into dedicated modules that: i) model interactions within and across wavelet sub-bands, depth features, and semantic priors, ii) apply semantic gating to the high-frequency bands, and iii) fuse modalities through an invertible coupling mechanism that prevents collapse onto a single modality. Extensive experiments across multiple benchmarks demonstrate that WAVE matches or outperforms existing methods, with the largest gains at high upsampling factors, where low-resolution depth contains the least structure.

Symbolic regression (SR) seeks to discover parsimonious mathematical laws from observational data, yet conventional approaches often struggle with the vast combinatorial search space of physically meaningful expressions. We present InsightSR, a framework that embeds Large Language Models (LLMs) as a guiding layer around the PySR genetic programming engine. Rather than relying on LLMs to generate expressions directly, InsightSR uses LLMs to progressively transform the search space itself through two complementary pathways: a Semantic Seed Pathway that proposes dimensionally consistent functional skeletons, and a Structural Feature Pathway that recommends nonlinear feature transformations. These transformations accumulate over iterations, broadening the input space and shifting the symbolic search from constructing deep expression trees over raw variables to assembling shallow trees over a rich, semantically informed feature set. A post-generation feedback loop evaluates candidates, categorizes features by their empirical utility, and refines the guidance for the next iteration, transforming the discovery process from open-ended generation into iterative, self-correcting refinement. Across three benchmarks, InsightSR achieves a 95% exact recovery rate on the Feynman benchmark and 80.18% accuracy on the LLM-SRBench LSR-Transform task, substantially outperforming state-of-the-art genetic programming and neural-symbolic methods while maintaining strong out-of-distribution generalization on real-world datasets.

Partial manipulation of speech recordings, where only localized segments of an utterance are altered, poses a significant challenge for content integrity verification, as reliable detection and localization of such edits becomes harder as the manipulated proportion decreases. Watermarking offers a proactive defense alternative by embedding auxiliary information prior to distribution; classical hash-based schemes achieve near-perfect detection and localization under ideal conditions, but the original content cannot be recovered once a segment is manipulated. Building on a prior self-embedding audio steganography framework, this work presents an initial exploration of proactive defense performance under ideal conditions, extending the investigation along three axes: frame-level localization, multi-bit least significant bit variants, and evaluation across multiple ultra-low-bitrate neural codec representations. By embedding a compact neural codec representation rather than a cryptographic hash, the framework additionally enables recovery of the manipulated regions, while supporting training-free detection and localization without spoofed examples. Experiments across four controlled manipulation types under ideal channel conditions show that the embedded payload, and hence an approximate reconstruction of the authentic content, is always fully recovered without bit errors. The results also indicate that the choice of neural codec is the dominant factor for detection and localization performance.

Artificial intelligence (AI) promises to accelerate biological research by automating computational analyses. Yet the ability of AI agents to carry out computational biology at the scale of complete research studies has not been systematically evaluated. Here we introduce BixBench3, a benchmark that measures the capacity of AI agents to process raw biological data through to scientific results. We designed BixBench3 tasks to mirror the delegation of work from a scientist to an agent: the scientist chooses the research question and high-level methods, then delegates implementation of all analyses to the agent. In each task, an agent receives a research objective, methodological guidance, and raw data derived from a published scientific study, and must execute a sequence of analyses to achieve the research objective. The data artifacts resulting from these analyses - such as peak call matrices or differential expression tables - are programmatically graded against the corresponding artifacts generated and reported in the original study. Across 20 BixBench3 tasks encompassing the generation of 138 unique artifacts, we find that 13 frontier models achieve scores ranging from 0.00 for Gemini 3.1 Flash Lite to 0.48 for GPT 5.6 Sol. Agents perform worse on tasks with larger raw datasets (0.36 on tasks with <100 GB versus 0.10 on tasks with >100 GB) and on analyses requiring more sequential steps (0.36 at 1-2 steps vs 0.24 at 3+). On average, agents use 6.8 hours, 102 million tokens, and $43 to complete each task, with the longest attempts consuming 24 hours, 1.07 billion tokens, and $525. Notably, the highest-scoring agents used fewer tokens and were cheaper than less performant options. These results reveal that LLMs vary substantially in their ability to (1) execute multiple sequential analysis steps coherently, (2) manage large quantities of raw data, and (3) work across scientific domains.

Partial deepfake speech, where only limited segments of an utterance are synthesized or manipulated, poses a significant challenge to existing deepfake detection systems. As the proportion of spoofed regions decreases, passive detectors become increasingly unreliable, and accurate detection and restoration remain challenging. In this paper, we revisit audio steganography from a new perspective and propose its use as a proactive defense against partially deepfaked audio. In particular, we consider a self-embedding strategy in which a clean speech signal embeds a compressed representation of itself, enabling post-hoc extraction of reference content. We demonstrate how existing audio steganography methods can be repurposed to support detection of partial deepfakes through codec-based restoration. Experiments on a benchmark dataset show that the proposed approach complements passive defenses. Remarkably, the proposed method operates without any training, providing a robust and data-efficient alternative for partial deepfake detection.

Physical human-robot collaboration requires a robot to provide assistance when human intention is clear while remaining compliant when several future motions are plausible. We present an adaptive stiffness framework based on generative action-chunk sampling. Conditioned on an RGB image and external joint-torque estimates, the policy samples multiple future action chunks from an observation-conditioned prior. Variation among the sampled action chunks is used to continuously adapt joint stiffness and damping. Greater variation makes the robot more compliant to facilitate human guidance, whereas lower variation provides firmer assistance. In a real-world collaborative transport task with four possible directions, the proposed method achieved an average success rate of 0.95, compared with 0.83 for a fixed-stiffness ablation and 0.69 for a deterministic baseline. Near direction determination, variation among the sampled action chunks increased and the controller accordingly reduced stiffness. These results suggest that variation among actions sampled by a generative policy can serve as an online control signal for balancing assistance and compliance in physical human-robot interaction.

Mixed-Integer Linear Programming (MILP) is a fundamental optimization paradigm in combinatorial optimization and has been widely applied across real-world domains. Due to its NP-hard nature, obtaining optimal solutions for large-scale or highly constrained MILP instances remains computationally prohibitive. Learning-based solution prediction has therefore emerged as a promising approach to provide high-quality variable assignment for solver acceleration. However, existing methods typically adopt a one-shot prediction paradigm that predicts the marginal probabilities of all variables simultaneously. As a result, the conditional dependencies among variables are only implicitly captured through message passing, with the burden of modeling the combinatorial structure falling entirely on the representational capacity of graph neural networks. To address this limitation, we propose the Structure-Aware Hierarchical Solution Prediction (SHSP) framework that replaces the parallel marginal decoding of one-shot methods with a novel hierarchical conditional decoding mechanism. Specifically, SHSP constructs a variable coupling graph from the constraint structure, decodes variables sequentially along a hierarchy of increasing coupling strength, and conditions each hierarchy on previously predicted assignments. To mitigate error accumulation during the decoding process, SHSP further incorporates a confidence-aware mask-and-repair mechanism to identify and correct unreliable intermediate predictions. We integrate SHSP with multiple learning-guided search methods, and evaluate it on four standard MILP benchmarks. Experimental results demonstrate that SHSP significantly outperforms existing one-shot prediction baselines, achieving a 54% average reduction in solution gap.

Multi-agent LLM systems coordinate through natural-language messages that consume 40--60\% of their token budget. Replacing these with structured graphs reduces cost but fails on tasks requiring adaptive reasoning. We propose \textbf{Routed Graph Handoff}, where a lightweight LLM router (155 tokens, 0.15\% overhead) selects between a typed dependency graph and natural language for each delegation. On four benchmarks (1,050+ trajectories), the routed system matches or exceeds NL-only on every task: \textbf{+12.7\,pp} on $τ$-retail at 3.2$\times$ compression ($p{<}0.01$), \textbf{+8.7\,pp} on BrowseComp at 2.2$\times$ compression ($p{<}0.05$), and parity on BFCL and AppWorld. Without the router, graph-only delegation regresses 14.6\,pp on AppWorld; the router eliminates this at near-zero cost. A graph-aware executor prompt is required: the same schema without interpretation guidance yields no gain. An oracle analysis reveals 8.6\,pp of additional headroom, motivating execution-time adaptive routing as future work.

Mixture-of-Experts (MoE) architectures enable scalable and efficient large language models (LLMs) by selectively activating expert sub-networks through a routing mechanism. However, this adaptive design introduces a new attack surface: specific experts become disproportionately correlated with certain tokens (e.g., end-of-sequence), allowing adversaries to manipulate model behavior via lightweight perturbations. In this work, we present \textbf{Groundhog Bit-Flip Attack (GBFA)}, the first bit-flip-based \textit{ Denial-of-Wallet availability attack} against MoE-based LLMs. By identifying and flipping routing-layer bits associated with related expert activations, we demonstrate that GBFA substantially extends the decoding token usage across three different LLM modes: conversational, reasoning, and agentic tasks, while largely preserving semantic fidelity. Across four main real-world MoE-based LLMs, manually deactivating on average fewer than \textbf{4 experts} drives average output inflation to $\mathbf{5912\%}$, with the majority of test samples reaching max tokens. These results reveal a robustness vulnerability of MoE architectures to bit flip, and highlight the potential of GBFA as an availability attack against LLMs.

Learned traffic-behavior models are commonly trained separately for each intersection, creating model portfolios that cannot share evidence across sites. We present PhaseShift, a topology-aware framework that harmonizes heterogeneous roadside trajectories into a shared actor-centric representation and trains one reusable backbone. Ego-relative coordinates, trajectory-induced movement paths, normalized signal context, and variable-cardinality interaction tokens remove site conventions while preserving behaviorally relevant topology. The backbone supports pooled operation, zero-shot at a held-out intersection, and low-data adaptation. We evaluate five intersections in two Florida regions on balanced field data, 100k training windows and equal-sized test sets per site under a replay-conditioned, best-of-sampled-trajectory protocol. At 10s, one pooled model lowers both minADE and minFDE relative to trained local models at all five sites, with median reductions of 36.8% and 22.0%. Leave-one-intersection-out deployment, including one cross-region fold, beats local training on both 10-s metrics at four of five sites, although short-horizon performance is less uniform. Fine-tuning with 1,000 target update windows improves on zero-shot at three sites and is the strongest regime at one. At site 7, every cross-site mixture sharply lowers long-horizon error under a fixed 100k-window budget; test-likelihood gains argue against a best-of-sample dispersion-only explanation. Local models fall behind calibrated IDM at the two highest-flow sites after long autoregressive rollouts; pretrained-backbone regimes do not. Within this five-site evaluation, PhaseShift demonstrates consolidation across heterogeneous physical control settings while identifying sites that still require adaptation. The protocol measures conditional single-vehicle generation under replayed context, not closed-loop traffic simulation.

Large language models (LLMs) frequently exhibit \emph{sycophancy}: they adapt their answers to a user's stated beliefs or preferences instead of reporting what they hold to be true, which lowers factual accuracy and can amplify misinformation. This paper proposes a methodology for mitigating sycophancy that employs the Bayesian Truth Serum (BTS), a peer-prediction mechanism, as the reward in Group Relative Policy Optimization (GRPO) to fine-tune an LLM. BTS pays an answer for being \emph{surprisingly common}, that is, more frequent among respondents than those respondents themselves predicted. We treat a group of responses from a model for one question as those respondents, so the reward is a function of the model's own outputs and fine-tuning needs neither labels nor preference annotations. We prove that in the large-group limit a sycophantic response earns strictly lower expected reward than an honest one. We also prove that if the entire group agrees in advance on a symmetric answering rule, it cannot earn a higher information score than under truthful reporting. On our true/false benchmark the reference model's answer-flip rate under user pressure decreases from 23% to 4%, and its accuracy under that pressure increases from 80% to 93%. Our reward outperforms SMART and is comparable to synthetic-data fine-tuning and to pinpoint tuning, all three of which train on labels. It spends considerably more compute in exchange, which makes it suitable when labeled data is scarce. Peer Truth Serum, which also pays a premium for a rare answer but elicits no prediction report, reproduces the effect. A peer-prediction reward computed inside a single GRPO group therefore reduces sycophancy without labels, and comparing mechanisms suggests that the premium paid for a rarer answer drives the effect.

Mixture-of-experts (MoE) models combine specialized predictors under learned routing, offering a principled mechanism for leveraging heterogeneity in medical data. We present a hierarchical multimodal MoE for interstitial lung disease (ILD) classification that integrates a frozen, pre-trained imaging expert with structured electronic health records (EHR) via two-stage gating. A modality-level gate assigns patient-specific weights to imaging and EHR predictions, while a sub-gating module decomposes the EHR branch into clinically defined feature groups with learned, group-specific contributions. This design preserves stable imaging representations while enabling input-dependent clinical weighting and explicit EHR specialization. Under strict patient-level cross-validation, the model achieved the highest mean AUC among the evaluated methods (0.8750 +- 0.0443), compared with 0.8646 for imaging-only REN and 0.7685 for SwinUNETR. The framework extends interpretability across anatomical regions, imaging--EHR utilization, and clinically defined EHR feature groups.

Conventional discrete choice and machine learning models are estimated primarily from observational data and typically treat explanatory covariates as parallel inputs, providing no internal mechanism for determining how related attributes should adjust when one is deliberately changed. This paper proposes Neural-Bayesian Structure Learning (Neural-BSL), a framework coupling differentiable structure learning with random-utility-based discrete choice estimation in a single differentiable procedure. To prevent mutually exclusive choice outcome from distorting the recovered attribute structure, the observed choice is maintained outside the graph as an alternative-specific utility comparison, while the attribute structure and random-utility parameters are learned jointly. The learned structure enters the choice model through structure-weighted attribute interactions and provides the structural basis for propagating interventions through downstream attributes. An intervention is evaluated by updating the intervened attribute, propagating its model-implied downstream changes in topological order, and then recomputing utilities and choice probabilities. This yields both predicted mode-share responses and the associated changes in downstream traveler or trip attributes. We evaluate Neural-BSL using stated-preference data from Seoul and the revealed-preference data from London. Neural-BSL achieves predictive performance comparable to conventional benchmarks while recovering behaviorally coherent dependency structures. Across policy scenarios, propagating interventions through the learned structure changes the predicted redistribution across modes while exposing the downstream traveler and trip adjustments underlying those responses.

Medical vision-language models (VLMs) can appear reliable in-domain while failing when acquisition domain, paired supervision, or evaluation protocol changes. We study this failure mode as a representation-level blind spot relevant to epistemic intelligence, without claiming a formal estimator of epistemic uncertainty. Using NIH ChestXray14 and CheXpert, we first isolate source-only cross-dataset visual transfer from unsupervised domain-adaptation diagnostics. Using PadChest and OpenI, we then evaluate multimodal alignment under strict pair-index retrieval and quantify metadata-derived source-proxy information retained in frozen embeddings. Self-supervised visual initialization improves NIH-to-CheXpert transfer over supervised ImageNet initialization in matched ResNet-18 comparisons, whereas adversarial adaptation is useful only in a narrow regime and becomes unstable as adversarial pressure increases. Multimodal exact-pair retrieval remains low under external OpenI stress testing, and source-proxy information remains recoverable from learned representations. Qualitative nearest-neighbor and Grad-CAM analyses show clinically plausible cross-dataset structure and thoracic attention patterns in many cases, while device-heavy and false-positive cases remain ambiguous. Auxiliary architecture checks are task-dependent and do not support a universal backbone ranking. Overall, the study shows that apparent competence under a single protocol can conceal transfer, alignment, and shortcut-related failure modes, motivating stress-tested evaluation of medical VLMs under distribution shift.

Badminton is a compact but challenging domain for game AI: a player must choose a physically feasible shuttle trajectory, anticipate the opponent's interception, and recover to a court position whose value depends on the opponent's next response. The central challenge is that shot selection and recovery are not separable: the best recovery depends on the shot-induced opponent response, while the value of the shot depends on whether the hitter can cover the reply. This paper presents ShuttleArena, a physics-based singles badminton self-play environment that couples continuous shuttle flight, player interception, structured shot generation, and post-shot recovery. The policy uses role-conditioned outputs: a masked interception choice on receiver turns and a factorized hitter action over shot azimuth, shot elevation, shot speed, and recovery target, enabling interpretable tactical probes. Episodes are single rallies rather than full scored games, and training uses Proximal Policy Optimization (PPO) self-play against a staged checkpoint opponent pool with sparse terminal rally-outcome rewards and a factor-specific recovery update. Evaluation with frozen checkpoint play, controlled tactical probes, recovery ablations, qualitative rollouts, and a human-data sanity check shows competitive improvement together with interpretable opponent-conditioned changes in shot geometry and recovery behavior. The learned policies produce recognizable badminton-like structure while also reflecting the abstractions of the simulator, and the recovery intervention shows that learned recovery behavior is competitively important. These results suggest that physics-based racket sports are a useful testbed for interactive digital entertainment AI because they require agents to coordinate execution, positioning, and opponent-relative tactical value.

LLM-generated search queries are widely used to augment IR evaluation, yet they may contain concepts that presuppose answer-side document knowledge, violating the information-access boundary of pre-search users. Existing validation metrics, including overlap, diversity, and effectiveness, cannot distinguish rare human-tail variation from candidate answer-side intrusion. We introduce concept provenance, a framework that assigns query concepts to backstory-supported, human-central, human-tail, and candidate answer-side zones, operationalizing a boundary that retrieval metrics alone cannot detect. Applying concept provenance to 77,004 queries across 100 UQV100 topics, 8 LLMs, and 5 prompt conditions with two extraction pipelines, we obtain a cross-pipeline token-HCIR Spearman rho of 1.0 over five condition means. Candidate answer-side concepts constitute 7.40 percent of non-generic concepts and appear in 97 of 100 topics, with topic explaining approximately 67 percent of variance. Human validation yields 68.2 percent relaxed precision, revealing two mechanisms: knowledge intrusion at 45.5 percent and deployment intrusion at 45.0 percent. Diagnostic probes show disproportionate localized retrieval effects, with deletion effect size d = -0.47 compared with d = -0.34 for random deletion, but these concepts explain less than 2 percent of aggregate evaluation variance. Concept provenance therefore serves as a boundary-compliance diagnostic rather than an evaluation-shift predictor. Under the tested conditions, no prompt condition eliminates intrusion; post-generation concept-provenance selection achieves 99 percent elimination.

Continual knowledge injection is essential for keeping large language models up-to-date in a fast-evolving world. Existing methods rely on supervised fine-tuning (SFT), which memorizes injected facts in their training format but fails to generalize across paraphrasing, document combinations, and reasoning. To address this, we propose Golden-GRPO Injection (GRIN), a three-stage self-learning framework for continual knowledge injection. Golden-GRPO is a mixed-policy reinforcement learning algorithm designed specifically for knowledge injection, which injects a golden answer to provide learning signal even when on-policy rollouts fail on novel facts. We further introduce Blank and Counter, two document-level benchmarks targeting novel acquisition and counterfactual overwrite respectively, each evaluating single-fact recall, multi-source retrieval, and inferential reasoning. Our experiments establish a clear empirical claim: mixed-policy reinforcement learning enables knowledge absorption beyond what supervised fine-tuning can achieve. GRIN substantially outperforms SFT and mixed-policy RL baselines on the harder question types while matching them on basic fact recall.

Coding agents increase development velocity but also technical debt. Prior work reports only average effects across adopters, hiding wide differences between teams. We introduce RAMP (Repository AI Maturity Profile), a four-level cumulative maturity model grounded in version-controlled artifacts that teams commit to configure AI tools. RAMP runs from behavioral rules and coding standards through named agent definitions to multi-agent orchestration, with observed practice concentrated in the first three levels. Across 441 repositories the levels behave as a cumulative scale, and independent human annotation reproduces RAMP's repository-level labels on 97% of a held-out sample. Adoption is cumulative, forward-only, and set-and-forget: 73.8% of artifacts are committed once and never modified. Re-estimating an existing agent-adoption panel within each stratum, agents accelerate development regardless of maturity (28-38% more commits), but quality diverges: among agent-first repositories, where the contrast is identified, those without committed AI configuration show roughly twice the increase in cognitive complexity (+53% versus +27%) and 1.7x the increase in static-analysis warnings. Because maturity is observational, correlated engineering discipline or model capability may explain part of the gap; we present these findings as hypothesis-generating and release RAMP as a reusable instrument.

In dynamic collaborative systems, the selection of reliable collaborators is critical to ensuring effective task execution. Existing trust evaluation methods often rely on unidimensional or scalar representations, which fail to faithfully capture a collaborator's true trustworthiness, thereby motivating a shift toward multi-dimensional trust modeling. However, due to the asynchrony of collected trust-related data across different dimensions, as well as the complex intra- and inter-dimensional dependencies embedded within these data, multi-dimensional trust evaluation remains challenging. To address these challenges, we propose TrustFormer, a task-specific multi-dimensional trust evaluation framework. Specifically, TrustFormer leverages task identifiers and device-generated timestamps to synchronize heterogeneous trust-related data across historical collaborations. It further employs cross-temporal and cross-dimensional attention mechanisms to jointly model temporal dynamics and inter-dimensional correlations, thereby effectively learning the multi-dimensional trust evolution of potential collaborators from historical performance data. In addition, according to the multi-dimensional resource requirements of tasks, potential collaborators' multi-dimensional resource trust is evaluated. Finally, by synthesizing these multi-dimensional trust profiles, the framework enables the optimal collaborator selection. Experimental results demonstrate that TrustFormer outperforms existing methods by yielding a 40.8% improvement in trust evaluation accuracy and enabling more reliable collaborator selection.

Clinical decision-making often involves prioritizing ethical values, such as beneficence, non-maleficence, respecting a patient's autonomy, and justice. Recent work has begun to assess how large language models (LLMs) make such subjective, value-laden clinical judgments. However, evaluations of LLM decision-making in rare disease care contexts, where ethical tensions are ubiquitous and where scarce prior information likely impacts LLM behavior, are still lacking. Here, we present a benchmark of 208 clinically grounded rare disease vignettes, each of which presents genuine, high-stakes conflicts. When prompting 11 state-of-the-art LLMs to choose between clinically defensible yet ethically conflicting next steps embedded within these vignettes, we found that all evaluated models consistently prioritized justice over other core bioethical principles. Specifically, models overwhelmingly favor equal resource allocation over need-based considerations, indicating LLMs' limited responsiveness to differences in clinical severity or situational context. We also identify a strong authority-framing effect: models favor justice in committee-based contexts and shift toward beneficence and autonomy only when final decisions are framed as being made by clinicians or patients respectively. Our work suggests that institutional pressures surrounding rare disease resource utilization may be silently reflected in LLM-based decision support systems, with finer ethical considerations disregarded.

Selection of trustworthy collaborators in distributed systems is critical for efficient task completion, necessitating the inference of trustworthiness from their past collaboration experience. However, as a collaborator serves distinct devices across diverse scenarios in past collaborations, its trust-related data, observed from different device-specific views, is inherently multi-source, heterogeneous, and uneven in quality. Consequently, achieving accurate trust evaluations for collaborator selection remains a major challenge. To tackle these issues, we propose a novel multi-view evidential learning (MVE) based trust evaluation method. First, to accommodate the multi-source heterogeneity of observed trust-related data, we model each task owner who has interacted with a potential collaborator as an independent observational view, enabling the evaluation of the collaborator's view-specific trust. Second, to address the dynamic evolution of trust under changing conditions, we leverage the powerful long-sequence modeling capability of the Mamba model to capture the deep temporal patterns of a collaborator's trust state within each view. Furthermore, to quantify the certainty levels of view-specific trust assessments, we incorporate an evidential deep learning mechanism in MVE, which outputs trust evaluation results while quantifying the subjective uncertainty underlying them. Finally, we employ a dynamic evidential fusion strategy to adaptively integrate the multi-view evidence based on their respective quantified uncertainties, thereby yielding a final trust evaluation for the collaborator. Extensive experiments demonstrate that the proposed MVE method outperforms baselines in both trust evaluation accuracy and task success rate.

Effective selection of trustworthy collaborators is crucial to ensuring the successful completion of collaborative tasks, which requires accurate assessments of both long-term device behavior and short-term collaborative dynamics. Consistent device behavior patterns, which are learned from historical collaborations, can be used to predict their reliability in future collaborations. However, accurately assessing device behavior based on historical collaborations remains challenging. First, behavior assessment from limited historical collaborations captures only instantaneous past behavior, failing to represent the devices' true behavior. Second, due to the temporal dependencies of device behavior, a unidirectional evaluation that relies only on earlier collaborations loses the opportunity to learn from subsequent collaborations. Addressing these challenges requires evaluating device behavior based on long-term collaborations while considering both forward and backward temporal dependencies. To this end, this work proposes a bidirectional Mamba-enabled model (BM) for long-term behavioral evaluation. For each short time slot, a graph is constructed among devices based on historical collaborations, and device behavioral features within the slot are then aggregated accordingly. Subsequently, a bidirectional Mamba model integrates these short-term representations across all time intervals, producing a stable and reliable long-term behavior evaluation for each device. Experimental results demonstrate that BM achieves higher evaluation accuracy than baseline methods, thereby enabling the selection of collaborators that maximize the value of task completion.

Mixture-of-Experts (MoE) models appear to encode moral content as robustly as dense models, yet prove far more fragile in their encoding. In OLMoE-1B-7B, linear probes recover moral valence from nearly every expert-layer combination, with mean peak-layer accuracy above 90%. But these representations collapse under levels of activation noise that a dense model of matched size easily tolerates, with a 4.2-fold difference in robustness. We trace this to output dilution. Because the MoE block averages across active experts before contributing to the residual stream, the feedforward signal reaching downstream layers is nearly two orders of magnitude smaller than in a dense MLP. Moral information, our interest, survives aggregation intact but at a scale trivially overwhelmed by perturbation. Routing itself remains stable under noise while the vulnerability originates entirely in the diluted aggregate. Checkpoint trajectories confirm this is architectural, not learned. Experts never specialize and accuracy saturates within the first few thousand steps. In sparse architectures, redundant encoding does not imply robust encoding.

Every deployed sparse-attention or KV-cache-eviction rule keeps a subset of the keys, discards the rest, and renormalizes the attention weights over the kept set. Enumerating the exact best subset under that constraint on $168{,}192$ attention rows from five models shows that keeping the largest weights is already near-optimal, since the best subset closes only a median $2$ to $5\%$ of the remaining gap to full attention. If selection closes this little, published margins between eviction methods must come from elsewhere, so we measure the bytes each method holds. In the shared evaluation pipeline, the strongest query-agnostic methods hold the full cache because their per-head selections are stored as masks, and only ragged per-head storage frees that memory. Enforcing a nominal budget on one fixed selection costs $14$ to $62$ benchmark points. We trace an $87.6$-point retrieval margin to rankings computed while the question is visible. ContourKV, a training-free allocator built from the dropped-mass statistic, wins $93$ of $160$ paired comparisons against that state of the art and loses $22$ at the byte count of the budget-enforcing baselines, and it ties the strongest of them.

We study exact representations of $\mathrm{MAX}_N(x)=\max{x_1,\ldots,x_N}$ using two-hidden-layer ReLU neural networks. This problem has been studied in recent years in an attempt to characterize the exact number of hidden layers required to represent continuous piecewise linear functions. The best lower bound is 2, while the current upper bound is logarithmic in $N$. It remains completely open if the right answer is a constant number of hidden layers (possibly even 2!) or not. In fact, a recent breakthrough was the representation of $\mathrm{MAX}_5$ as a two-hidden-layer ReLU function obtained in [Bakaev et al., 2026], and the case of $\mathrm{MAX}_N$ was stated as open for $N\geq 6$ in that paper. Using a careful computer assisted search, we obtain two-hidden-layer ReLU representations of $\mathrm{MAX}_5, \mathrm{MAX}_6, \mathrm{MAX}_7$, and $\mathrm{MAX}_8$. We obtain these by considering rational linear combinations of terms of the form $\max\{\sum_{r=1}^{s}\max(x_{a_r},x_{b_r}),\sum_{r=1}^{s}\max(x_{c_r},x_{d_r})\}$, where $a_r,b_r,c_r,d_r\in\{1,\ldots,N\}$. Each inner maximum of two coordinates can be computed in a first hidden layer, and the outer maximum of the two side-sums can be computed in a second hidden layer. Consequently, every finite linear combination of these terms has a two-hidden-layer ReLU realization. An identity for $\mathrm{MAX}_N$ in this form therefore gives an exact two-hidden-layer ReLU representation of $\mathrm{MAX}_N$. Very recently, two-hidden-layer representations of $\mathrm{MAX}_N$ of the above form were obtained for all $N\leq 10$ in [Ruess et al., 2026]. Our representations are different and were developed independently. While our techniques share most of the high-level ideas presented in [Ruess et al., 2026], there are also some minor differences which may be of interest for future research on this problem.

Mixed-Integer Linear Programming (MILP) is a fundamental tool for combinatorial optimization with extensive real-world applications. A central challenge is designing computationally efficient MILP formulations. Large Language Models (LLMs) offer new opportunities to automate the modeling process, from deriving formulations to strengthening them. Reliable automation requires robust methods for verifying that proposed formulations preserve the underlying optimization problem. However, existing approaches evaluate formulations numerically and fail to reason about general problem instances. We resolve this limitation by introducing a constructive definition of MILP reformulation that can be formalized in Lean and machine-checked. We develop FLARE (Formulation-Level Automated Reformulation Evaluation), a method that uses an LLM-based agent and the Lean proof assistant to verify proposed reformulations against a reference formulation. To evaluate our approach, we introduce FormulationBench, a challenging dataset of 20 problems and 109 formulations. FLARE outperforms existing methods, with 100% accuracy on the NP-hard subset of FormulationBench. Furthermore, FLARE produces a machine-checkable certificate for every reformulation it accepts. For cases where formal guarantees are not necessary, we introduce FLARE-NL, a fast and cheap LLM proxy that matches FLARE's accuracy but produces no certificate. These methods enable reliable verification in automated optimization modeling.

Speakers in natural conversation take turns speaking and listening, deciding in real time when to take, hold, or yield the floor. However, turn-taking evaluation remains limited due to the lack of a consistent, linguistically grounded evaluation protocol and hand-annotated data covering diverse conversation types. To address this, we present TurnBench, a multi-domain benchmark that pairs a 30-hour, hand-labeled corpus of dyadic human conversation with a standardized evaluation protocol for end-of-turn and interruption detection. We set conversation type as a controllable experimental variable, covering six distinct interaction styles, and triple-annotate each conversation. Benchmarking 14 heterogeneous turn-taking systems, we find end-of-turn recall stable across types, while interruption false positives are strongly type-dependent and concentrated in backchannel-dense interaction styles. Although in smooth floor transfers human listeners begin speaking a median 151 ms before the current turn ends, no current system performs equivalently without incurring excessive false positives. We release our corpus, a 104-hour training set, and a public leaderboard with an interactive dataset viewer at https://turnbench.sesame.com

We present an LLM-driven, datasheet-aware framework for early-stage hardware compatibility verification that identifies documentation-level interface incompatibilities based on hardware datasheets and high-level component connectivity descriptions. It does not require, and can therefore be used, before detailed schematic simulation and implementation. We view trustworthy LLM-assisted design automation not as directly generating answers from documents, but as transforming engineering information through traceable verification stages. Given hardware datasheets and high-level component connectivity descriptions, the framework constructs a design graph that captures device connectivity and shared interaction domains, retrieves only the engineering properties required by explicit, domain-oriented verification criteria , and generates deterministic scripts for compatibility evaluation. By decomposing compatibility analysis into modular stages and preserving intermediate results, the framework reduces context overhead, improves transparency and tractability, enables scaling, and avoids reliance on LLMs for numerical computation. Evaluated on seven embedded-system designs comprising 34 datasheets, our framework achieves 97.5% compatibility-verification accuracy and an 8.6 times reduction in input context size compared with ``upload-and-query'' workflows. These results demonstrate the feasibility of LLM-assisted, specification-based hardware compatibility verification at an early design stage, as well as the need for, and substantial benefits of, modular task decomposition, formalized verification criteria, and task-aware compact context construction.

Natural language driven autonomous co-scientist workflows involve a fundamental trade-off between flexibility and reasoning at the expense of determinism, reproducibility, and observability. Such agents increasingly must communicate across institutional boundaries, where federation topology can shape latency and cost. We systematically evaluated these tradeoffs using a controlled ablation on a production agentic platform for science. We use a verifiable task: given a protein sequence, we ask an agent to confidently characterize its function by routing across common tools. We compare federation topology, classic RL vs LLM-driven harnesses, language model, and prompt expertise. We also stratify results by protein novelty. We find that the choice of LLM dominated prediction quality far more than topology or prompting (Opus ~92%-94% vs o4-mini ~40%-50%). The PPO policy was nearly as accurate as the best LLM (88%) at zero token cost, fastest latency, and perfect consistency, but yields no reasoning trace. Expert prompted LLMs reached the highest accuracy but were high-cost and less consistent; prompt dependence was largest when the task was hardest. Federation imposed a negligible penalty on performance. These results offer actionable guidance for deploying agents for scientific workflows: for routine, verifiable tasks, a cheap deterministic policy delivers near-frontier accuracy with complete reproducibility, while flexible LLM reasoning is best reserved for open-ended discovery.

We introduce LibriBrain100, a large-scale MEG dataset for speech decoding designed from the ground up for reproducible, standardised evaluation. LibriBrain100 more than doubles the size of the original LibriBrain release, resulting in over 100 hours of high-quality MEG acquired while subjects listened to naturalistic continuous speech. With $\sim$80 hours from a single subject, LibriBrain100 sets a new record for deep, within-subject neural data (8$\times$ more than the next comparable dataset and roughly 80$\times$ more than other datasets). To demonstrate the payoff of this depth-first design, we evaluate on a word-classification benchmark---an increasingly well-established stepping stone towards the open challenge of noninvasive brain-to-text decoding. Using an existing decoding model, we achieve state-of-the-art performance---validating both the quality of the recordings and the value of within-subject data at scale. Because collecting 80 hours of data per user is impractical for real-world applications, we also collected $\sim$40 minutes of additional data from each of 32 subjects. Using the same word-classification benchmark, we demonstrate the value of broad multi-subject data: supervised finetuning of a pre-trained model can substantially compensate for limited per-subject data. We provide standard train, validation, and test splits, all reproducible through an open-sourced Python library that supports easy downloading, optional preprocessing, and data loading for common deep learning frameworks. In addition, the dataset and evaluation infrastructure are being released alongside an open machine-learning competition with a public leaderboard for standardised benchmarking. Ultimately, our hope is that LibriBrain100 will accelerate progress towards practical non-invasive brain-computer interfaces, capable of restoring communication to people living with severe paralysis.

Spec-Driven Development (SDD) is a fast-emerging practice in which a structured natural-language specification, written by a developer, or (more often) drafted by an AI tool and then curated by the developer, drives an AI coding agent's implementation. A wave of tooling (GitHub Spec Kit [3], OpenSpec [4], AWS Kiro [5], and dozens of others) has appeared since 2025, yet the artifacts these tools produce have never been studied at scale. We present SPECMINE, a corpus that captures SDD in public GitHub repositories through two censuses: a broad census of spec.md/specs.md files covering most tools (470,795 files across 73,030 repositories, attributed to 17 named tools), and a Kiro census of its distinct requirements/design/tasks layout (98,574 files across 12,910 repositories). Each spec is enriched with full repository metadata, complete commit history, and parsed document structure. How a spec becomes code is itself an open question, so for 11 tools we sweep every pull request that touches a spec in their repositories with at least ten stars, capturing 5,992 such PRs across 581 repositories with their changesets. That makes the simplest workflow, spec and implementation changing together in one PR, directly observable, and a census-wide index of 2,421,323 typed references (1.28M to code files, 863k to sibling documents, 152k to PRs, 62k refs, 43k branches, 22k issues) gives a second, independent link from spec to code. SPECMINE lets the community study, for the first time, how software is specified in the age of AI agents.

We consider the problem of learning a mixture of $k$ Plackett-Luce models given multi-way ranking responses from annotators that may represent heterogeneous underlying preferences. This problem has many applications in AI alignment and preference optimization. Prior work has studied mixtures of Bradley-Terry models from pairwise comparisons. However, uncovering mixture models is theoretically unidentifiable when $k$ exceeds $m/2$, where $m$ is the length of a ranking. We propose an efficient implementation to address this limitation, which involves first augmenting the rankings to a larger size by generating new responses from a base language model, followed by a gradient-based estimation to reduce inference cost in the input embedding space. Based on this procedure, we then design an expectation-maximization algorithm with these two steps to fit a mixture of Plackett-Luce models, called MoPLEx. Extensive experiments are conducted to verify this approach. First, we show that the gradient-based approximation estimates true probabilities with less than 5% error on models with up to 34 billion parameters. Second, we show that MoPLEx improves clustering and ranking accuracy by an average of 43.7% and 15.2% over baselines using single ranking and mixtures of Bradley-Terry models, on preference optimization datasets. These results demonstrate the effectiveness of MoPLEx for tackling multi-way rankings from heterogeneous preferences through measuring alignment between gradients.

We study a tree-structured regularizer over class-prototype layouts in a hierarchical-classification model and ask whether the choice of latent manifold for the prototypes (Euclidean R^d vs. the Poincare ball B^d_c) affects how well that regularizer can be satisfied without distorting the data likelihood. The two manifolds differ only in their volume growth: hyperbolic space grows exponentially with radius and embeds trees with provably lower distortion than R^d of matched dimension, so the structured regularizer should be cheaper to satisfy on B^d_c. Across 150 seed-replicated regularized maximum-likelihood fits spanning embedding dimension, curvature, and regularizer strength on WikiArt (27 styles, 81,446 paintings, frozen CLIP ViT-B/16 features), we find a single robust effect: Poincare prototypes preserve the topology of the nearest-neighbor graph in latent space substantially better than matched Euclidean prototypes (sibling recall@5 +8.7 pp, cousin recall +15.2 pp; paired-t p < 10^-4, sign agreement 0.94), and the gap holds across three reference-tree definitions (hand-built lineage, CLIP-derived, and DINOv2-derived). On classification, Euclidean prototypes are tied with logistic regression on raw encoder features, indicating no detectable contribution from the latent geometry; only the hyperbolic fit improves on a k-NN encoder baseline for local retrieval. Global tree-fidelity comparisons are unstable across reference trees and we do not claim a winner. The results give an empirical separation, on a real hierarchical-classification problem, between two natural latent geometries for a class-structured regularizer.

Deciding whether to call a tool is a core competence of an LLM agent, and a costly one to get wrong: needless calls add latency, accrue cost, and may trigger irreversible side effects, while missing calls leave the model confidently wrong on questions it could only answer through tool-calls. Models manage this balance poorly, both over-using and under-using tools. Existing methods such as post-training and prompt engineering are expensive and difficult to modify at inference time. We show that whether an instruction-tuned model calls a tool can be controlled by a single linear direction in its residual stream, extracted without any training from the model's own tool-use preference signal and turned into an inference-time intervention with no prompt change. Adding the direction with strength $α$ moves the call rate monotonically from near $0\% $ to over $90\%$ while keeping calls well-formed. The steering works in both directions: dialing it down suppresses calls, and dialing it up induces new calls that land precisely on the questions the model cannot answer from its own knowledge. We also show that the direction generalizes to unseen tools with strength comparable to each tool's own direction and without favoring any specific tool choice. With live tool execution, a single sweep of the steering traces a cost/accuracy Pareto frontier and nearly doubles open-domain QA accuracy ($0.29 \! \rightarrow \! 0.56$); the same recipe transfers across a diverse range of models spanning dense, MoE, and multimodal architectures, without any training. Our code is publicly available at https://github.com/YuqiChen4188/Steering-Tool-Use-Propensity.

Smart freight corridors offer a practical pathway for connected and automated vehicle (CAV) deployment in freight transportation, but physical experimentation is expensive and existing approaches rely on predefined control policies that cannot capture adaptive behaviors. This paper presents an agent-based modeling (ABM) framework coupling a physical infrastructure layer, a connectivity layer (V2X), and a decision layer integrating reinforcement learning (RL) and multi-agent reinforcement learning (MARL) for platoon formation and charging coordination. We evaluate three scenarios (Baseline, Assisted, and Cognitive) using throughput, congestion, energy, emissions, and robustness metrics. Preliminary results indicate that the Cognitive scenario achieves higher throughput and lower congestion than the baseline, while the Assisted scenario delivers meaningful energy savings per kilometer through platooning. Sensitivity analysis shows that the throughput advantage of the smart corridor widens under conditions with high demand and that MARL coordination extracts greater utilization from fixed charging capacity than rule-based assignment.

Fake news detection has become an important Natural Language Processing (NLP) task due to the rapid spread of misinformation through online news platforms and social media. While transformer-based models such as BanglaBERT achieve strong performance for Bangla text classification, their quadratic computational complexity makes them less suitable for long-document processing in resource-constrained environments. This paper investigates Mamba-based State Space Models (SSMs) as an efficient alternative for Bangla fake news detection. We propose BanglaMamba and compare it with pre-trained BanglaBERT and a similarly configured BERT model trained from scratch. Experimental results show that BanglaBERT achieves the highest Macro-F1 score (0.9260), while BanglaMamba (0.9029) achieves performance comparable to the from-scratch CustomBERT (0.9057) despite using a different architecture. Meanwhile, BanglaMamba achieves approximately $2.2\times$ higher inference throughput and 49% lower inference peak GPU memory usage than the BERT-based models. Cross-dataset evaluation shows that BanglaBERT generalizes better to an external dataset, highlighting the importance of large-scale pretraining. These findings demonstrate that Mamba-based SSMs can provide a competitive and computationally efficient alternative to Transformer-based architectures for Bangla fake news detection, particularly in resource-constrained settings.

Understanding how molecular interactions govern macroscopic behaviour is a central challenge in molecular sciences. However, conventional theory building cannot keep pace with the vast datasets modern experimentation routinely produces. Large language models offer a promising route to automating theory construction, but a spatiotemporal field cannot be directly placed in a prompt. Existing models generally learn about the data only through a score measuring how well each proposal fits it. Here we introduce data interpretation, a stage that measures the field into the quantities a theorist would consult and supplies them to the model as a direct input. On a benchmark of simulated fields, interpretation nearly triples the accuracy of recovered equations relative to showing the raw data, at negligible computational cost and without any training. By allowing a language model to read field data as a theorist does, data interpretation offers a practical route to automated field theory construction that can coevolve with experimentation.

Most competitive 4-bit LLM research pipelines now open the same way: apply a linear, function-preserving transform (rotation, scaling, permutation, non-orthogonal affine) so the outlier mass sits more favorably against the group scales, and only then round. Yet we are aware of no survey dedicated to this transform stage, and its literature is quietly re-deriving an older theory. We identify and formalize the principle that organizes it, the Great Inversion: allocation-flexible coding rewards energy concentration, whereas the grouped shared-scale quantization a deployed matrix instruction performs rewards within-group flattening. Classical transform coding (1963: decorrelate, allocate bits, quantize) spends different bits per coordinate at a fixed total rate; for a Gaussian source at high rate the Karhunen-Loeve transform's concentration minimizes distortion. A deployed operand tile instead carries one absolute-maximum scale per group and equal bits everywhere, with no allocation; on a uniform grid that objective rewards flattening, approached by Hadamard incoherence. We prove that opposition under within-group majorization: the prescriptions point in opposite directions, each backed by a proof against its own objective, and for a generic spectrum no optimality guarantee transfers. A second axis is the number format: the non-uniform FP4 grid makes flattening buy less, MXFP4's power-of-two block scale still rewards a rotation confined to that block, and NVFP4's mantissa-carrying scale largely removes that pull, so the target pole depends jointly on allocation regime and format. We survey 200 works to a June 2026 cutoff; classify 43 transform methods by structure, data-awareness, searched-versus-constructed, and runtime cost; record, where reported, how they compose with GPTQ rounding; distill a first-choice guide by deployment regime; and close with the open problems it exposes.

In this paper, we study alternating regret in online convex optimization (OCO), motivated by the success of alternating learning dynamics in two-player games. Although previous works have shown that $o(\sqrt{T})$ alternating regret is achievable under various assumptions on the loss functions and feasible domains, the minimax regret rate has remained open even for the expert problem. In this paper, we resolve this question by showing matching lower and upper bounds for both the expert problem and general OCO. Somewhat surprisingly, for the $d$-expert problem, we show that the minimax alternating regret is $Θ(\log d)$, independent of the horizon $T$. This significantly improves upon the best-known $\mathcal{O}(T^{1/3}\log^{2/3} d)$ established by Hait et al. [2025]. We further extend our results to general OCO over a $d$-dimensional compact convex set and prove that the worst-case minimax alternating regret is $Θ\left(d\log \left(1+\frac{T}{d}\right)\right)$, also significantly improving upon the best-known $\mathcal{O}((d\log T)^{2/3}T^{1/3})$ upper bound and resolving the open problem posed by Cevher et al. [2023], Hait et al. [2025]. Technically, our upper bound for the expert problem is achieved by a corrected variant of Hedge, in which carefully designed correction terms cancel the unfavorable curvature arising in the alternating-regret analysis. We extend the same corrected-potential argument to continuous action sets to obtain the optimal alternating-regret rate for OCO. For the lower bounds, the expert construction repeatedly eliminates half of the candidate experts, while the OCO lower bound instance construction replaces this discrete elimination by a more involved multiscale construction on the unit disk.

Collective dynamics arise in a wide range of physical, biological, and engineering applications. Examples include cell migration, swarm robotics, social dynamics, and animal behavior. A defining characteristic of these systems is the emergence of large-scale coordination from local interactions among agents; a fundamental question is thus to understand the local interactions that give rise to the observed emergent dynamics. We are interested in methods for learning interactions generally, which can describe a wide class of physical systems exhibiting collective dynamics defined by an interaction kernel, without a priori assumptions on the analytical form of this kernel (i.e. it is nonparametric). The advantage of this kernel-based approach is that it incorporates the underlying physics of the model (i.e. collective dynamics), which more general equation-learning approaches may ignore, potentially limiting their effectiveness for model accuracy and predictions. In this work, we extend existing variational learning approaches to collective systems with both interaction kernels and environmental/intra-agent forces. The proposed framework simultaneously infers the interaction kernel non-parametrically while learning the environmental force using either semi-parametric or fully nonparametric representations. The methodology is validated on several benchmark models exhibiting synchronization, alignment, attraction-repulsion, and external environmental forces. We also introduce a model-selection procedure based on our nonparametric learning framework to identify models that optimally explain a given set of trajectory observations. By exploiting the feature-identification capability of the learned models, the proposed procedure can distinguish among different collective dynamics frameworks and recover mechanistic interaction mechanisms directly from trajectory data.

Worked examples are a important part of introductory programming, but reading their expert explanations is passive. Self explanation, students explaining the problem and its solution to themselves with subgoal level analysis, turns that study into an active task, yet it is hard to scale because assessing free-text explanations and returning timely feedback has had no easy automated solution. We investigate whether a large language model (LLM) can fill that gap. We build a self-explanation tutor for introductory programming, ESSE, in which students explain lines of worked examples and receive immediate LLM feedback on the correctness and completeness of each explanation, and we pursue two goals. First, we ask whether the LLM judges student explanations well enough to serve as the engine of the tutor; we assess its judgments against two independent human reference standards of different kinds, a single domain expert and a crowd of non-expert raters, each with its own strengths and weaknesses, characterizing both where the LLM is reliable and the systematic tendencies in how it diverges. Second, we ask whether the LLM-based tutoring benefits students; deploying it in an introductory Java course, we find that its feedback leads students to persist and revise rather than abandon a line, that their explanations grow more complete and conceptually richer across attempts, and that students show evidence of learning. These indicate that LLM-based assessment is good enough to power a self-explanation tutor, and that the tutor positively shapes how students study worked examples.

Sidewalk-scale path extraction demands perception and planning that run reliably on compact, low-power hardware in cluttered, map-sparse environments. We present a monocular vision pipeline for sidewalk path extraction in micromobility systems that progresses through three design iterations, from a skeleton-graph baseline through distance-transform corridor planning to a lightweight image-space architecture, and provides a systematic comparison of five path-planning methods across both bird's-eye-view (BEV) and image-space domains. A compact SegFormer-B0 student model, trained with a semi-supervised teacher-student framework using OneFormer Swin-L pseudo-labels, achieves a hand-annotated IoU of 0.946 at 11.7 ms per frame, improving over the baseline checkpoint (IoU 0.758, 18.9 ms). In a controlled planner comparison on 32 hand-labeled frames, image-space midpoint planning achieves the lowest lateral center error (14.3 px) at 2.2 ms, a 421x speedup over BEV distance-transform planning (926.8 ms, 65.0 px center error), while maintaining comparable mask-path alignment (98.5% versus 98.6%). A full-video replay across six campus sequences (22,679 frames) confirms that the improved segmentation reduces temporal instability from 1.46% to 0.33% and increases template-path availability from 73.7% to 79.3%. We further show that BEV-only path extraction is fragile in monocular settings: in one profiled run, 99.3% of frames produced no valid BEV path. The final recommended architecture, image-space midpoint primary, image-space distance-transform fallback, and BEV reserved for visualization, runs the full perception-to-path stack in under 50 ms per frame on CPU, making it suitable for embedded pedestrian-speed micromobility systems.

Audio clustering is a fundamental task for organizing rapidly growing speech collections, supporting applications such as conversational analysis and speech-driven discovery. However, existing methods rely on fixed acoustic similarity metrics or ASR-based text pipelines, limiting their ability to reorganize the same audio collection under different user-specified perspectives, especially when clustering depends on both linguistic and paralinguistic cues. We introduce audio multi-perspective clustering, where a model directly partitions speech recordings according to a natural-language perspective while inferring both the number of clusters and their assignments. To study this setting, we construct AudioLens-Bench, a benchmark spanning multiple application domains and evaluating both in-perspective and cross-perspective generalization. We further propose AudioLens-R1, an end-to-end large audio-language model trained with reasoning distillation and preference optimization. Experiments show that AudioLens-R1 consistently outperforms all baselines, improving overall ARI by 12.99 points and V-measure by 11.62 points. These results demonstrate the promise of native audio-language models for flexible, perspective-conditioned structure discovery over speech collections.

Structural health monitoring (SHM) is essential in modern engineering, providing data for condition-based maintenance, lifecycle assessment, and predictive decision-making. Traditionally, SHM relied on visual inspection to detect defects such as cracks and deformations. Early computer vision (CV) methods, including thresholding, edge detection, and handcrafted features, aimed to automate this process but were highly sensitive to noise, imaging variations, and multiscale defects, limiting their reliability. Recent advances in machine learning, particularly convolutional neural networks (CNNs) and You Only Look Once (YOLO), have improved defect detection accuracy and enabled real-time analysis. However, adoption in SHM remains limited due to technical barriers such as data labeling, model training, and deployment, which typically require programming expertise. To address this gap, we introduce YOLOEZ, an open-source, GUI-based tool for end-to-end YOLO model application. YOLOEZ integrates data labeling, training, and inference into a single interface, enabling high-performance model development without code while supporting reproducible workflows. Evaluation against existing software and classical image processing demonstrates that YOLOEZ not only outperforms traditional methods across most detection metrics, but also lowers adoption barriers present in other modern CV tools. By combining accuracy with accessibility, YOLOEZ facilitates wider use of AI-driven monitoring for predictive maintenance, digital twins, and intelligent structural systems.

Coding agents increase implementation capacity without automatically making project intent, system structure, or acceptance evidence explicit. As implementation becomes abundant relative to engineering judgment, the scarce work shifts toward choosing useful abstractions, producing evidence, and determining which obligations govern acceptance. Existing workflows address parts of this gap through larger prompts, repository retrieval, or perchange review, but still require agents and engineers to reconstruct consequential properties. As an alternative, we present Model-Based Agentic Software Engineering (MAGE). MAGE is a framework and a theory for building trustworthy autonomy from commodity intelligence. MAGE addresses a representation problem and an authority problem: it externalizes the smallest purposeful representation needed to answer an engineering question, then gives settled obligations proportionate authority through constraints, sensors, validators, and gates. It keeps uncertain intent open and turns recurring reconstruction and judgment into durable engineering structure that later work can inherit. We developed MAGE from a longitudinal case and refined it through six independently reported industrial accounts. Across these sources, MAGE explains how externalized knowledge, bounded action, independent evaluation, and retained human authority can compose into a governed engineering environment, and proposes tests of when that environment turns commodity intelligence into durable engineering progress.

Transformer representations describe trajectories through high-dimensional vector spaces, which are shaped dynamically as tokens incorporate relational context across layers. Such data tend to concentrate on lower-dimensional sub-manifolds, a form of compression quantified by the Intrinsic Dimensionality (ID), the minimum number of independent variables needed to represent them without significant information loss. In this work, we ask whether the grammatical role of tokens, as marked by their part-of-speech (PoS) tag, shapes the local geometry of this manifold. To this end: (1) We investigate the layer-wise evolution of ID, finding that closed-class items expand earlier and collapse sooner than open-class ones; (2) We show its expansion and contraction to be explained by changes in the neighborhood structure, and hence in the relations between words within a sentence; (3) We compare encoders (ModernBERT, bigbird-roberta-large) and decoders (gemma-2-2B, Llama-3.2-3B), finding that the two families evolve differently across layers, consistently with how each integrates context;(4) We show that geometric features alone recover a token's grammatical role, and use them to interpret how the semantic content of each PoS evolves across layers in a downstream classification task.

Offline reinforcement learning (RL) leverages static datasets to learn decision policies without real-time environment interaction. While recent sequence-modeling approaches rely on continuous diffusion models for trajectory synthesis, applying these methods to discrete planning tasks requires a categorical formulation rather than the standard Gaussian construction. We present BFN-RL, a unified generative modeling framework for offline RL based on Bayesian Flow Networks (BFNs). By iteratively evolving distribution parameters rather than noisy data instances, BFN-RL natively models both discrete and continuous trajectory spaces within a single probabilistic formulation. The categorical planner generates future state sequences, and a learned inverse-dynamics model converts consecutive generated states into actions. Evaluations in discrete planning and continuous control show that BFN-RL can generate effective trajectories across both categorical and continuous state spaces. Our results establish BFNs as a versatile generative foundation for offline trajectory planning across data modalities.

Waveform inversion seeks to estimate the wave speed of a heterogeneous, inaccessible medium, from time-resolved measurements of the waves at user controlled sensors. We consider this inverse problem for acoustic waves and an active array of source/receiver sensors that emit probing signals and measure the generated pressure waves. The forward map, from the wave speed to the measurements, is nonlinear and oscillatory. The oscillations cause cycle skipping, the main impediment to using the standard, nonlinear least-squares data fitting formulation, known as full waveform inversion (FWI). A recently introduced alternative waveform inversion approach computes from the measurements an algebraic surrogate of the wave operator, a reduced order model (ROM) matrix, which is then used to estimate the wave speed. The mapping from the measurements to the ROM is nonlinear, but well understood. It is computed efficiently, in a non-iterative manner. The nonlinear mapping from the ROM to the wave speed is less understood, and its approximation involves time-consuming optimization. Our goal in this paper is to use a neural network to map the ROM matrix to a nearby one, that has a simpler and explicit dependence on the wave speed. This simplifies and reduces the computational cost of the ROM-based waveform inversion. We introduce the methodology, called ROMNet, and test it with numerical simulations, using two training data sets: The first set consists of random media with variations of the wave speed modeled by a superposition of Gaussians with random amplitudes and standard deviations. The second is the publicly available GeoFWI dataset introduced for benchmarking FWI using deep learning. We compare the performance of ROMNet with the direct ROM-based inversion and with two representative deep learning approaches to FWI: ``Fourier-DeepONet" and ``InversionNet".

Evaluating the ability of large language models (LLMs) to discover software bugs is increasingly important. Existing benchmarks typically evaluate this capability by asking the model to generate a proof-of-concept input that triggers a predefined target vulnerability. However, this setup may overlook valid crashes discovered by the model when they do not match the predefined target. As a result, the evaluation may not reflect the model's real capability. We present FuzzingBrain-Bench, a benchmark for assessing AI models' ability to discover bugs in open-source software. Models are given an open-source project and a sanitizer-instrumented harness in a self-contained Docker image. Their goal is to generate inputs that trigger as many distinct crashes as possible through the harness. A model's performance on each challenge is scored based on the number of distinct crash signatures it produces, capped at a predefined maximum and weighted by a difficulty coefficient. FuzzingBrain-Bench V1 consists of 77 challenges drawn from 43 open-source projects, with 36 C, 32 C++, and 9 Java/JVM challenges. We evaluate Claude Haiku 4.5, Claude Sonnet 4.6, and Claude Opus 4.8 on the full benchmark. Claude Opus 4.8 performs best, triggering crashes in 60 of 77 challenges and achieving a score of 196 out of 579. None of the three models triggers a crash in 13 challenges. The FuzzingBrain-Bench corpus and harnesses are publicly available at https://github.com/fuzzingbrain/FuzzingBrain-Bench.

Multi-agent LLM systems increasingly debate answers, coordinate research, simulate users, and mediate information flows, making agent-to-agent persuasion a basic but undermeasured capability. We introduce a controlled testbed for studying how goal-directed persuaders shift elicited stances in networks of LLM agents grounded in real-world ego-network topologies. Across four LLM backbones, five graphs, and 55 policy statements, we find that persuasion dynamics depend on the interaction between topology, competition, topic, and model prior. Additionally, we show that direct exposure reliably predicts next-round stance change in competing runs, and peer relays carry smaller but measurable influence, showing that agents not assigned to persuade can still transmit persuasive force. Finally, analyzing post text alone misses important movement: planned strategies are only partly realized in executed messages, action choices can diverge from message content, and persuadees rarely state the stance shifts detected by probes. These results argue for evaluating multi-agent persuasion as a trajectory- and exposure-level process, using belief probes, exposure provenance, and action logs to identify who influenced whom and whether visible language reflects underlying stance movement.

Knowledge graph (KG) construction pipelines must continuously integrate newly arriving entities into a growing graph. Unlike inserting triples between existing nodes, a newly arriving entity has no graph connectivity: it emerges from the acquisition phase as a raw feature vector and must be assigned to a semantic community before entity resolution and link prediction can operate over a tractable candidate set. Existing multi-view graph clustering methods exploit multiple relation types as structural views, but are transductive: they assume a fixed graph and cannot assign unseen entities without retraining. We propose SNAP-KG (Streaming Node Assignment via Projection for Knowledge Graph Entity Integration), a framework supporting graph-structural multi-view relational clustering and inductive inference for streaming entities. SNAP-KG trains a projector to map a new entity directly to the learned embedding space using only raw features, enabling immediate cluster assignment without graph access or model retraining. Experiments on five benchmark multi-view graph datasets and a production-scale KG of 2.4 million nodes demonstrate multiple orders-of-magnitude inference speedups over retraining-based approaches and competitive clustering quality. As a candidate scoping mechanism for downstream tasks, SNAP-KG achieves 62-75% candidate search reduction on the five benchmark datasets and 97% on OGB-WikiKG2 for entity resolution and link prediction.

Frozen hematology foundation-model (FM) embeddings reach near-saturated in-domain white-blood-cell (WBC) accuracy, but clinical deployment demands reliability across scanners, sites, stains and preparation pipelines. We audit 15 frozen encoders (hematology, pathology, and general vision) across four public single-cell acquisition domains along two axes: accuracy robustness and calibration. In-domain linear-probe macro-F1 is saturated (0.98-0.997), yet cross-dataset macro-F1 drops 34-72% and rankings re-order: DinoBloom-L, the in-domain best, falls to 10th of 15 on the most-shifted target (MLL23) at the benchmark's shared 224-px input, behind RedDino and several general and pathology encoders. Rank transfer is probe-dependent: 1-NN retrieval is more stable on average than a source-fitted linear head (median $ρ$ 0.65 vs 0.45), but neither probe universally predicts target robustness. Calibration also collapses: source-trained probes are nearly calibrated in-domain (expected calibration error, ECE, 0.004) but confidently wrong off-domain (ECE 0.35), and source-fitted temperature scaling transfers poorly. We further audit pretraining exposure and identify MLL23 as DinoBloom's internal cohort; because DinoBloom's only held-out dataset is also our source domain, this benchmark cannot isolate exposure from scanner-associated shift. Label-free adaptation and marginal-entropy-based model selection appear safe under balanced evaluation but fail under realistic WBC class-prior shift. Class-Balanced Re-standardization (CBR), a training-free pseudo-label-balanced feature normalization, improves all evaluated target-prior scenario means and partially improves calibration, although encoder-level exceptions and residual miscalibration remain. Hematology FM benchmarks must therefore jointly audit accuracy, calibration, exposure, and class-prior robustness.

Autonomous driving has made remarkable progress through imitation learning with massive human demonstration data. However, a trained planner often degrades severely when applied to a new environment zero-shot, because of domain shifts in traffic regulations, road layout and driving behaviors. Therefore, adapting a trajectory planner to a new city typically requires resource-demanding local data collection with a vehicle sensor suite. In this work, we show that driving behavior can be learned from a scalable and efficient alternative. We introduce \emph{SkyDrive}, a framework that utilizes drone-based traffic monitoring to provide efficient supervision for autonomous driving agents in a new environment. While vehicle-based data collection logs the ego and its surroundings, an aerial platform naturally observes many road users simultaneously over an extended field of view. As a result, every vehicle can be a data source with grounded driving behavior, effectively scaling up the amount of supervision. Based on 137 hours of aerial traffic monitoring footage, we extract 650K driving samples and construct a benchmark for trajectory planners and motion predictors. Zero-shot experiments with multiple models reveal significant cross-city domain gaps, but many of them can be alleviated by limited supervision from the sky, e.g., 30 minutes of monitoring per location. Our findings show that aerial traffic monitoring is an efficient and scalable data source for adapting autonomous driving systems in new cities. Data and code will be made publicly available.

Existing approaches to building line-level Arabic handwritten-text-recognition (HTR) training data either rely on fully manual annotation, which does not scale, or on automatic OCR-to-reference alignment methods not yet extended to multi-script, two-zone (main-plus-margin) manuscript layouts with a provable correctness guarantee. We present RefLAM (Reference-grounded Line Annotation for Manuscripts), a pipeline converting manuscript page images and clean transcriptions into validated, line-level ground truth without sacrificing human oversight. RefLAM couples a deep-learning page-segmentation model with a multimodal large language model (MLLM) for structured OCR and a diacritic-agnostic fuzzy alignment engine that grounds each OCR line in a contiguous span of the reference text, with a character-level confidence score in $[0,100]$. A perfect score is provably equivalent to character-for-character identity of the normalised strings (the Confidence-100 rule), verified with no counterexample across the released corpus. A reviewer can thus trust a perfect score, confirming most lines at a glance rather than retyping them, so annotation becomes triaged, with attention concentrated on uncertain alignments. Across 7 fully page-validated books we measured a 75$\times$ throughput gain over manual annotation (3,000 vs. 40 lines/hr); applying the same guarantee to 7 further books, we retained 16,533 confidence-100 main-text lines within one week, excluding sub-100 lines rather than manually correcting them. Using RefLAM, we release AraMS-28k: 14 historical Arabic manuscript books, 3,043 pages, and 27,971 main-text and 629 margin-line annotations with bounding boxes, layout labels, and insertion anchors for 191 margin entries (30.4%). We also finetune Muharaf-pretrained baselines (including HATFormer) on AraMS-28k and report CER results confirming its practical utility for downstream HTR training.

Stochastic masking, cropping, or modality removal makes deterministic reconstruction an incomplete target: one observation can admit many clean completions. This work takes the corresponding posterior $P(X\mid C)$ as the common statistical object for conditional generation and generatively sufficient representation learning. Drift Variation autoencoder trains a masked encoder $Z=E(C)$ and a conditional flow decoder with one clean-prediction Flow Matching loss. The analysis first decomposes the ideal conditional KL into generator approximation and the representation deficiency $I(X;C\mid Z)$. It then derives orthogonal risk decompositions for conditional Flow Matching. For an affine Gaussian path, the clean-prediction representation gap is zero if and only if $P(X\mid Z)=P(X\mid C)$. Thus the encoder-dependent excess clean-prediction risk induced by Flow Matching and the profiled ideal conditional KL have the same posterior-sufficient zero set, without being numerically equal objectives. An exact conditional field with a zero-noise endpoint then generates $P(X\mid Z)$ and hence $P(X\mid C)$ at a joint ideal optimum. The result extends to continuous multimodal product spaces when the complete modality tuple remains the Flow target for every observation mask. On CrossGeom-4, an 18-run controlled benchmark, observable factors have linear-probe $R^2$ of $0.9990$-$0.9992$, shuffling the joint model's encoder condition increases conditional error by $13.5\times$-$15.7\times$, and joint target attention reduces disagreement on an unobserved factor shared by two outputs by $90.1$-$92.8\%$ relative to independent target decoders. Visible modalities are also generated and reconstructed, directly validating the full-tuple objective. Unconditional mode balance remains imperfect, delimiting the empirical claim to a controlled multimodal proof of concept.

The development of federated learning (FL) techniques has helped improve the privacy preservation of users' data and extended the applications of machine learning models. However, the involvement of a large number of users in FL also creates open opportunities for different adversaries, such as poisoning attacks, Byzantine attacks, and adversarial example attacks. Yet, recent research has disclosed that existing poisoning attacks and Byzantine attacks can not achieve satisfactory penetration in realistic FL scenarios caused by strong assumptions, \textit{e.g.,} client selection rate, and the ratio of malicious attackers. In this paper, the transferability of adversarial examples among different client models is analyzed to understand the relation between adversarial examples and clients' data distribution. Moreover, to mitigate the attacks of transferable adversarial examples, we design a defense mechanism stemming from the transferability of model robustness by adversarial training. As a result, through theoretical analysis of transferability, we gain insights into adversarial examples and the vulnerability of federated learning systems. Our proposed adversarial attack and defense methods are evaluated via real-life datasets in various settings to show their performance over the existing state-of-the-art methods.

Multi-modal time series forecasting methods integrate auxiliary context into temporal predictions through increasingly sophisticated fusion mechanisms. A growing body of work reports substantial gains, yet it is often unclear whether they reflect genuine use of the context or incidental architectural effects. We ask a narrower, checkable question: when can auxiliary context help a forecaster at all? We identify two dataset-level conditions that must both hold: (1) the target is not dominated by a last-value shortcut (low autocorrelation rho_h), and (2) the context carries information about the target beyond history (non-zero conditional mutual information delta; when delta=0 no predictor can benefit---a distribution-free result). Through controlled experiments on MoME (a 14.3B-parameter mixture-of-experts model, 6 datasets, 10 seeds) and four additional fusion mechanisms implemented within a single-backbone testbed (5 datasets), we find that when both conditions hold, text-conditioned expert modulation contributes a sizeable MSE reduction; when either fails, the contribution collapses to the capacity floor of the modulation pathway and carries no context-attributable signal. We establish causality through two interventions: adding a shortcut to MoME suppresses routing contribution by 77-93% across 3 datasets; progressively corrupting context quality drives the context-specific benefit from +44% to negative. We validate the autocorrelation component of our diagnostic on 27 Monash Archive datasets. We provide a calibrated pre-training diagnostic that, on the datasets we test, yields no false positives in well-powered settings. We are explicit about the asymmetry of our evidence: the negative arm is broadly reliable, while the large positive magnitudes come from a single model family (MoME) and are corroborated only in direction by the testbed.

Machine learning has had a significant positive impact on the prediction of athlete performance and injury risk. Most works in this field rely on subjective observations and expert assessments, which restrict their effectiveness. In sports like soccer, basketball, and wrestling, some studies attempt to address this challenge by integrating data from alternative sources, such as readings from wearable devices, alongside traditional subjective observations and expert assessments to enhance accuracy. However, similar research in tennis remains largely unexplored. In this paper, we propose a multimodal Predictive Athlete Readiness framework for Tennis (PART) to assess both performance and injury risk in tennis players. By leveraging machine learning and deep learning techniques, PART processes multiple sources of data collected from nine collegiate tennis players, including physiological metrics, training and match data, sleep data from wearable devices, self-reported information via daily questionnaires, jump assessments, and motion analysis from match play videos. PART captures four characteristics of tennis players: overall wellness, injury risk, physical capability, and playing style. By integrating these four characteristics by supervised learning, it is capable of providing a holistic assessment of the tennis athlete's condition, along with advanced forecasts of specific body areas at risk such as the upper body (e.g., elbows) or lower body (e.g., knees). Our evaluation, conducted with data from nine collegiate tennis players, shows that PART achieves strong performance in predicting both overall wellness and injury risk. Additionally, our framework also shows promise for recreational tennis players, who often suffer from injuries due to incorrect playing techniques.

Retrieval-augmented generation (RAG) improves large language models by incorporating external knowledge without retraining, but existing methods often underuse the relational structure encoded in knowledge graphs. Graph-based RAG can capture entity relationships, yet supervised graph retrieval typically requires labeled question-answer data that may not be available for newly constructed graphs. We address this limitation with SelfGraphRAG, a framework that generates question-answer pairs directly from knowledge graph structure and uses them to train a query-conditioned graph retriever. The generated questions capture multi-hop paths and local neighborhoods, providing relational supervision without manual annotation. Experiments on multi-hop question answering and classification benchmarks show that SelfGraphRAG improves retrieval precision and downstream reasoning performance over embedding-based baselines. These results suggest that knowledge graph structure can provide useful supervision for training graph retrievers when labeled data are unavailable.

Machine learning is usually formalized through samples, while the persistent individual to which multiple observed or possible events refer often remains implicit. We propose the \emph{unit} as an explicit primitive at the level of task semantics. A learning task first declares a population of persistent referents and a sameness criterion; the realized value $u$ denotes the selected referent. Supervised learning is the main formal specialization. Its semantic object is a family of unit-conditioned response laws. Homogeneity is the special case in which those laws coincide; a sample-only conditional is silent as to whether the world is homogeneous or the observed law is only the marginal of a heterogeneous family. What is learned from data is a pair $(T_φ,R_θ)$: a tokenizer that produces a contextual unit token and one shared response-law form that reads it. The structured class takes that form to be a simple relation in the token; a linear predictor is the running instance. The token is the learner-side representation through which the task-side unit affects prediction, while a learner specification that omits unit information is unit-insensitive; homogeneity remains a property of the world-side response family. When identity is unresolved, the world-side law mixes unit-conditioned targets, while the learner composes its shared form with a token. A trusted resolver may fix the unit and supply a lookup token; otherwise \emph{unit abduction} forms a token of the same type from factual evidence. Unlinked single-row observations can fail to distinguish a heterogeneous unit world from a homogeneous pooled world; trusted same-unit pairs separate a restricted witness. The formal results concern this supervised specialization.

Optimizing expensive, high-dimensional black-box functions remains a central challenge in modern machine learning and scientific discovery. While local Bayesian optimization mitigates the curse of dimensionality, existing techniques often prioritize the probability of descent over the magnitude of progress. This leads to overly conservative steps that yield negligible improvement, wasting queries on directions that are nearly certain to descend but offer little decrease. We introduce Gradient Refinement and Progress-Aware Exploitation (GRAPE), a two-stage framework that first sharpens the local gradient posterior via a closed-form acquisition function, then selects update directions by maximizing the expected decrease conditional on descent. Theoretical analysis proves that this gradient refinement stage monotonically minimizes local uncertainty and that the progress-aware direction converges to true steepest descent as the posterior sharpens. Empirically, GRAPE demonstrates superior query efficiency across high-dimensional tasks: in black-box adversarial attacks, it achieves an average 5.4$\times$ speedup over baselines, and on large language model prompt optimization tasks, it outperforms the second best method by a reduction of 3.8 log-units in the final average regret.

Existing methods for improving Retrieval-Augmented Generation (RAG) efficiency mainly optimize downstream LLM generation, such as context compression or serving optimization. However, RAG is an end-to-end system, and its bottleneck can shift between upstream reranking and downstream generation under different serving loads and reranking budgets.In this paper, we first empirically characterize this shifting-bottleneck behavior and show that upstream reranking can become the dominant bottleneck under high query rates or large reranking budgets. Reducing the reranking budget can relieve this bottleneck, but it may also drop supporting evidence and degrade recall. To address this problem, we propose \textbf{\textsf{PACE}} (\textbf{P}rioritized \textbf{A}daptive \textbf{C}overage of \textbf{E}vidence), a training-free framework that combines \textit{evidence frontloading} with \textit{pressure-adaptive budgeting}. \textsf{PACE} first reorders candidates by marginal evidence coverage, prioritizing documents that are query-relevant, complementary, and useful for forming multi-hop evidence chains. We show that this objective is monotone submodular, giving greedy selection a $(1-1/e)$ approximation guarantee. \textsf{PACE} then dynamically adjusts the reranking budget according to the relative pressure of the reranker and the LLM. Experiments on three multi-hop QA datasets and online serving simulations show that \textsf{PACE} improves evidence recall, reduces p95 latency under ranking-heavy workloads. More importantly, the two components together reveal that \textit{less can be more}: an evidence-dense top-ranked candidates enable higher final recall with fewer reranked documents.

Federated learning (FL) has emerged as a key approach for training models across decentralized data, yet benchmarking in FL remains difficult to reproduce, compare, and extend. Existing evaluations are often tied to custom infrastructure, released as incomplete research code, and conducted primarily in simulation, which limits portability and practical relevance. We present Flower Hub, a platform for publishing, discovering, and executing decentralized and federated applications. We show how it enables reproducible benchmarking by packaging benchmarks as executable, versioned applications with standardized metadata, pinned dependencies, and explicit evaluation workflows. We instantiate this approach with a multi-domain benchmark suite spanning cross-silo and cross-device settings, and including tasks in medical imaging, financial tabular learning, legal instruction tuning, phishing URL detection, and audio tagging. We further demonstrate that the same benchmarking application can run across both simulation and deployment runtimes without changing the application code, enabling unified evaluation across varying learning environments. Beyond model quality, our benchmark design supports system-aware reporting, including runtime and communication metrics. This work advances benchmarking in FL settings from ad hoc code artifacts towards portable, executable, and reusable benchmark applications.

Large Language Models (LLMs) are powerful but limited by static parametric knowledge that becomes outdated once pretraining ends. Knowledge editing addresses this problem by updating model behavior on target facts without full retraining. In particular, in-context knowledge editing has gained attention because it is training-free and readily applicable to black-box LLMs. Recent reinforcement learning (RL)-based approaches improve over fixed retrieval strategies by adapting prompt construction to the quantity-quality trade-off. Despite initial success, they fail to model the prompt as a structured entity under the distinct and often competing objectives of reliability, generality, and specificity. Previous methods largely optimize a single objective and make decisions over only part of the prompt construction process, thereby overlooking both the balance of different objectives and the global organization of demonstrations. We propose Multi-Objective In-context Knowledge Editing (MO-IKE), a multi-objective RL algorithm that formulates prompt construction for in-context knowledge editing as a Constrained Markov Decision Process. MO-IKE trains a dynamic retriever to optimize competing objectives in knowledge editing, enabling more balanced and globally coherent prompt construction. On Llama-3.2, MO-IKE improves edit success (reliability) from 85.0% to 92.0%, paraphrase consistency (generality) from 77% to 79%, while increasing retention rate (specificity) by 23.0% compared to prior RL-based methods.

Understanding how (multimodal) large language models perform on physics problems requires benchmarks that reflect the difficulty and breadth of expert-level physical reasoning. Existing physics benchmarks remain limited in the following two important ways: (1) short of high-difficulty datasets, and (2) lack of comprehensive coverage of visual forms, knowledge points, and step-by-step solution processes. As a result, model performance on current datasets may not be fully representative of their ability to solve complex physics problems. To address these issues, we present PhysElite, a large-scale bilingual multimodal benchmark for Olympiad-level physics reasoning. PhysElite contains 11,586 Olympiad-tier problems. For each problem, we provide corresponding visual diagrams, step-by-step bilingual Chinese-English solution derivations, and the final answer. We benchmark 18 open-source and closed-source MLLMs, and find that even the strongest model reaches only 33.7% answer accuracy. We additionally conduct step-level process evaluation to diagnose where models fail in the reasoning chain. Our datasets are released at https://huggingface.co/datasets/physelite/PhysElite.

The growing adoption of large language models (LLMs) has raised increasing concerns about the energy consumption and environmental impact of inference. This paper presents a systematic empirical study of decode-phase energy consumption across representative open-source LLMs employing Multi-Head Attention (MHA), Grouped Query Attention (GQA), and Grouped Query Attention with Sliding Window Attention (SWA) to characterize how attention architecture influences decode-phase energy consumption under varying inference workloads. We evaluate four models across different context lengths, batch sizes, and generation workloads while measuring GPU energy using NVIDIA hardware counters. We examine the effects of context length, attention mechanism, Key-Value (KV) cache growth, and batching on decode-phase energy consumption. Results show that attention mechanism is the primary factor governing how decode energy scales with context length. MHA models exhibit substantially steeper energy growth than GQA models, whereas GQA with SWA maintains nearly constant energy consumption. We further show that model size primarily determines absolute energy consumption, while batching reduces both energy per generated token and request latency by up to 87%. These findings provide practical guidance for selecting energy-efficient LLM architectures and inference configurations.

The proliferation of distributed energy resources (DERs) in distribution grids enables the active coordination of these assets to reduce costs and enable cleaner operations. Realizing this potential requires solving multiphase AC optimal power flow (AC-OPF) quickly across varying loads, DER availabilities, and topology reconfigurations, at much greater speed and scale than conventional nonlinear solvers. Learning-based surrogates can offer millisecond inference, yet existing methods target largely balanced transmission systems and do not scale to the multiphase, unbalanced, and reconfigurable nature of distribution feeders at utility scale. We present the Penalty + Sequential Linearized Feasibility Seeking (SLFS) algorithm, a self-supervised learning framework for multiphase distribution AC-OPF under switch-induced topology changes. Penalty+SLFS requires no labeled optimal solutions and trains directly from the AC-OPF objective and constraints through a differentiable fixed-point power flow solver, avoiding expensive label generation and admitting robust training procedures. Topology changes are handled efficiently using Sherman-Morrison-Woodbury updates of the admittance-matrix inverse, while an M-step Jacobian approximation accelerates differentiation through the power flow solver. At inference, SLFS repairs any infeasible predictions, providing feasibility guarantees with low computational overhead. On IEEE feeders ranging from 13 to 8,500 nodes, Penalty+SLFS achieves negligible optimality gaps and near-zero constraint violations, delivers up to three orders of magnitude speedups over IPOPT, and remains robust under large distributional shifts, demonstrating a viable path toward real-time, topology-aware AC-OPF for large-scale distribution grids.

Self-evolving Skill harnesses (AutoSkills, Hermes Agent) generate more advisory orchestration automatically; their reported gains are efficiency, not safety. This misses the actual gap: a Skill describes how an agent should behave; a Policy decides which behavior is allowed to become an action. Today's format covers the first with markdown and scripts; the second is left to the model. Generating more Skills scales the gap, not the safety, especially when a wrong invocation can unlock a door or move money. Two adjacent attacks are documented: malicious skills compromising cloud software, and jailbroken LLM-controlled robots causing physical harm. Their intersection, malicious agent skills causing physical harm, follows directly but has not been reported. We name this class Borrowed Authority: Skills format gives the receiving agent no typed way to reject an inter-agent permission claim, so a malicious or misused Skill can drive actuation by attaching one. We propose Edge Skillguard, a typed authority layer that lives inside the Skill artifact rather than between tools as workflow engines do, with guards over world state and sensor evidence. On a live edge control-plane testbed, the guards reject 60/60 borrowed-authority requests across five attack variants without blocking benign requests, and the result holds at 5x scale and across hosts over a Tailscale mesh. These results suggest that high-risk Skills should co-package typed invocation policy with procedural knowledge, so that physical actions depend on machine-checkable evidence rather than peer-agent claims.

Crosslingual evaluation of language models that enables fair comparisons remains a fundamental challenge in multilingual NLP. Existing studies adopt a variety of downstream tasks and intrinsic metrics with different theoretical justifications, yet there has been little empirical investigation into whether these approaches yield meaningful crosslingual conclusions. We systematically examine crosslingual evaluation approaches using controlled monolingual language models trained on parallel data with varying tokenizer vocabulary sizes and model sizes, and further validate our findings on multilingual LLMs. We further discuss challenges in achieving comparable downstream evaluation across languages. Our results show that several widely used normalized metrics introduce crosslinguistic biases rooted in tokenization, encoding, and orthographic differences. In contrast, sentence-level negative log-likelihood computed over semantically equivalent sequences provides more meaningful and consistent crosslingual comparisons.

Cortical computation is strikingly low-dimensional: a handful of latent variables, carried in a neural population's activity, steer the higher-dimensional responses of individual neurons. Our aim is sample efficiency-models that decode well from limited data and at small parameter budgets. In a standard Transformer layer, the feed-forward block applies the same operator to every token. We suggest a von-Neumann inspired hypothesis of efficient computation as an alternative for neural decoding: a controller decodes an instruction and then executes a token-specific operator; the usual realization-a soft mixture of experts-only blends their outputs, not operators. We introduce a von-Neumann State-Space Transformer (VN-SST), a memory-augmented Transformer whose feed-forward block is a low-rank instruction bank: a shared base operator plus a small set of learned low-rank instructions, from which a per-token code synthesizes the weight matrix actually used at that token. The code is read from a low- dimensional projection of a carried state-space memory, so a slow latent trajectory acts as an instruction pointer-mirroring how low-dimensional dynamics may route cortical computation. On three motor-cortex neural-decoding benchmarks, VN-SST is far more data-efficient than a modern Transformer, each jointly predicting spikes and decoding behavior. This model wins by a wide margin on the scarcest benchmark, leads on the other two, and turns longer context into rising rather than falling accuracy. We evaluated that the network compresses a large instruction bank to a few bits per token, so program capacity acts as a control channel, not an accuracy lever. The same model is also more parameter-efficient on two small text benchmarks used for language modeling (LLMs), suggesting a generic mechanism.

Clinical diagnosis is fundamentally interactive and incremental, yet the dominant paradigm for evaluating Large Language Models (LLMs) in medicine remains static QA benchmarks or template-based dialogues. These benchmarks say little about whether a model can serve as a diagnostic agent in a dynamic clinical encounter, with LLMs showing significant accuracy and reliability degradation in multi-turn settings. To address this issue, we present MTDiag, a large multi-turn diagnostic dialogue dataset constructed from three heterogeneous sources: DDXPlus, MIMIC-IV, and published case reports (AJCR), covering common ED presentations as well as long-tail rare and atypical conditions. All cases are normalized into a canonical schema anchored in the most comprehensive and widely-adopted medical knowledge bases (UMLS concept identifiers, with ICD-10 diagnosis codes). We release the schema, a UserLM-8B-based utterance-generation pipeline, and the physician-validated dataset that converts structured clinical evidence into natural-language utterances. Importantly, we introduce and motivate clinical knowledge-grounded metrics for evaluating LLMs as diagnostic agents, beyond diagnostic accuracy, for the task of multi-turn differential diagnosis.

We present a natively multiscale operator learning method for the surrogate modeling of (numerical solvers for) multiscale partial differential equations (PDEs). The primary novelty of our method lies in a novel multiscale kernel frame function approximation technique. Leveraging this new kernel frame technique, we cast the operator learning problem as one of learning frame coefficients of output functions as a function of frame coefficients of input functions. The generalization step then automatically allows for a multiscale decomposition of the output functions. Our method is applicable to both tensor-product grids and point clouds. We present interpolation proofs, error estimates, and numerical convergence rates for our frame approximation. We the demonstrate the applicability of our method for the surrogate modeling of inherently multiscale PDEs. The new multiscale frame kernel method is significantly more accurate than popular neural operators on challenging problems from the literature, while simultaneously admitting an a posteriori multiscale decomposition upon generalization.

Time series forecasting models are widely used in high-stakes settings, yet their predictions remain difficult to interpret because existing post-hoc methods often ignore temporal dependence and fail to provide horizon-specific explanations. We propose a model-agnostic explainability framework that explains forecasting predictions by attributing each forecast horizon to temporally relevant historical lags. The framework models forecasting as a latent trajectory and introduces semantic flow to quantify how information evolves across time in the model's internal representations. By aggregating semantic flow, it constructs a lag-horizon attribution matrix that captures horizon-resolved temporal influence. To improve explainability, we further generate structure-preserving perturbations and fit sparse local surrogate models, producing human-readable and temporally coherent explanations. We evaluate the method using faithfulness and stability diagnostics across multiple benchmark datasets. Results show that the semantic-flow variant achieves competitive or superior faithfulness compared to standard post-hoc baselines, while being substantially more computationally efficient. Stability analysis further demonstrates that the explanations are robust and identifies regimes where interpretation should be applied with caution.

Digital mobility outcomes (DMOs) derived from wearable sensors characterise mobility in daily life and offer a promising means of monitoring disease progression. Yet most DMO studies examine one disease at one visit; they do not model how multivariate DMO relationships with multiple clinical outcomes evolve jointly across diseases. Technically, existing temporal multi-task frameworks can model progression within an individual disease, but they do not jointly model multiple prediction outcomes across diseases, particularly when disease cohorts do not share participants. To address these gaps, we propose DeMMO, an interpretable framework for longitudinal, multi-disease, and multi-outcome learning. DeMMO represents each disease-outcome objective by a longitudinal DMO coefficient matrix and combines temporal regularisation with stable and visit-specific feature selection. Its central technical contribution is an automatic cross-disease and cross-outcome relation-learning mechanism that learns signed relations directly from these longitudinal mappings, enabling selective information sharing without paired participants. We evaluate DeMMO on the recently released, large-scale, multicentre Mobilise-D dataset, which provides a new opportunity to study 24 harmonised real-world DMOs over five visits across multiple mobility-limiting conditions. Against nine strong linear, longitudinal, and deep-regression baselines, DeMMO achieves the best overall and outcome-specific prediction performance, with significant improvements over the strongest baselines. Stability selection further identifies reliable longitudinal DMO patterns for subsequent clinical validation and disease monitoring. The implementation code and experimental results are available at https://github.com/menghui-zhou/DeMMO.

General-purpose health benchmarks increasingly anchor claims about LLM medical performance, but they are not always resolved by clinical specialty, making domain-specific performance hard to isolate. Mental health is of acute public-health concern as millions of people turn to LLMs for psychological support, and most existing evaluations are bespoke academic benchmarks that are difficult to integrate into developer workflows. We introduce HealthBench-Psych and HealthBench-Psych-Hard. We screened HealthBench's 5,000 physician-rubric conversations for mental-health relevance with a transparent LLM-applied rubric, then validated the subset through two rounds of blinded clinician review with concealed known-exclude controls, yielding 610 conversations (12.2% of the corpus). Evaluating 20 frontier and open models under a cross-vendor panel of three LLM judges, we find a statistically tied frontier cluster, measurable refusal behavior in two models, and near-identical rankings across judges ($τ\ge 0.92$). We release the subset, pipeline, model responses, grades, and analysis code as a reusable resource.

Simulated evaluation is widely used to benchmark AI agents, yet how much evidence a simulated pass provides about physical deployment has not been systematically quantified. We present SimVerity, a verdict-transfer assurance framework: it replays matched scenarios on target smart home deployments and cross-validates agent execution against independently qualified physical witnesses. Our evaluation highlights that deployment success is a real-world process, not a static property in simulation: completion, reported state, observable effect, and settled outcome diverged within the same execution. Although an advanced simulator cleared all 240 light trials, a camera caught 42 sub-second failures invisible to settled-state checks. False clearance was predictable: a risk profile learned from measured trials and locked before evaluation predicted failures on a path it never physically measured, beating a property-blind baseline in all eleven held-out sessions across two cohorts. Agent auditability was also measurable: switching one agent loop's model-client/serving configuration raised its scenario-matching share from 52-88% to 100%. Finally, a second qualified simulator added no independent cross-check: it never disagreed on any overlapping case, and only physical measurement exposed their shared blind spots. SimVerity turns verdict transfer into an explicit decision: clear, abstain, or escalate before deployment.

LLM inference is increasingly constrained by accelerator memory capacity rather than compute throughput. This constraint is especially acute in single-accelerator and small-node inference systems, where limited on-package memory capacity restricts the size of deployable models. HBF is an emerging 3D-stacked NAND flash technology that provides multi-terabyte near-accelerator capacity, making it a promising capacity tier for storing LLM weights. However, existing HBF-based proposals face three adoption challenges: they (1) rely on coarse-grained static prefetching for LLM weights aiming to hide the microsecond-level read latency of the NAND flash device while maximizing HBF's read throughput, (2) expose NAND flash management tasks (e.g., refresh operations) to the accelerator-visible critical inference path, and (3) miss optimization opportunities to specialize and optimize the flash-management mechanisms to the workload behavior. Our goal is to design an efficient HBF substrate that integrates HBF as a memory-capacity tier alongside HBM while addressing these three challenges. To this end, we propose FLINT, a workload-driven HBF substrate for capacity-scalable LLM inference. FLINT introduces three mechanisms: (1) a hardware burst-buffer controller that dynamically coalesces and pipelines HBF reads aiming to utilize existing NAND flash buffers while sustaining high HBF bandwidth, (2) a phantom-plane refresh mechanism, which removes refresh from the critical inference path by moving refresh-related NAND flash operations outside the read foreground back via low-cost resource duplication, and (3) a read-only FTL, which replaces SSD-class support for arbitrary writes with a compact table that translates logical weight bursts to physical HBF locations.

GPUs increasingly accelerate database systems, but query-specific peak performance still often relies on hand-written kernels. Existing LLM kernel benchmarks focus on machine learning operators, leaving irregular, heterogeneous, data-movement-heavy database-style operators untested. We introduce DataKernelBench, which translates SQL into validated PyTorch TorchPlan programs and evaluates LLMs that optimize either the core tensor-bounded snippet or the full query in CUDA or Triton through execution-guided repair. Across ten proprietary and open-weight models on TPC-H SF10 with an H100 GPU, the strongest full-query CUDA configuration achieves $2.11\times$ speedup over torch.compile at full pass rate. We find that higher-performing implementations commonly use kernel fusion and execution-strategy changes, stronger models benefit most from full-query specialization, and workload context matters more than hardware context. To handle data larger than GPU memory, we extend TorchPlan with Dask-cuDF for on-demand partition loading on TPC-H SF100 with four H100 GPUs, achieving $2.54\times$ speedup

Fully few-shot class-incremental audio classification (FFCAC) requires recognizing new sound classes from only a handful of labeled examples per session, without forgetting previously learned classes and without any large base dataset. Existing methods typically freeze a pre-trained audio--language encoder and classify with point prototypes, but they suffer from significant performance degradation throughout the sessions due to generic feature representations. We propose SPECTRA, a framework built on a frozen encoder which adds three components. (i) a lightweight trainable adapter that calibrates the generic embeddings to the task; (ii) subspace feature replay, an exemplar-free anti-forgetting scheme that replays old classes by sampling from the low-rank subspace of their stored features; and (iii) a transductive optimal-transport refinement of prototypes at test time. Our central finding is that the subspace structure of the replay diminishes forgetting and outperforms naive Gaussian replay of equal variance. On three FFCAC benchmarks (NSynth-100, FSC-89, LS-100), SPECTRA improves average accuracy and reduces forgetting over current state-of-the-art methods, and our ablations statistically validate each component.

Edge LLM deployment is shaped by more than model size and precision: inference backend, hardware platform, memory traffic, and power management all affect latency and efficiency. We present Hydra, a common-schema, phase-aware workload characterization framework for LLM inference on edge SoCs. Hydra instruments HuggingFace Transformers and llama.cpp with a shared per-prompt timing schema and fuses those records with hardware telemetry, enabling a multi-dimensional characterization of performance, system-resource utilization, and efficiency across prefill and decode phases. Using Hydra, we evaluate three consecutive edge System-on-Chip (SoC) generations (AGX Xavier, AGX Orin, and AGX Thor), 13 instruction-tuned LLMs from seven families, five execution formats, and consider input/output-length sensitivity. The resulting artifact contains roughly 107K per-prompt records and is publicly released with Hydra. Our analysis shows that aggregate latency alone hides key deployment effects: backend structure changes where latency is introduced, quantization reduces memory traffic and energy but does not predict power monotonically, and SoC generation changes how utilization and efficiency should be interpreted. By connecting phase-level timing with system-resource utilization and efficiency metrics, Hydra enables reproducible, phase-aware characterization of edge LLM inference. Hydra's source code and the collected per-prompt trace corpus are available open-source at: https://github.com/amirtaherin/hydra

Hessian-free high-resolution (HFHR) dynamics augments underdamped Langevin dynamics (ULD) with reversible position diffusion for sampling problems that arise in machine learning. We establish an explicit quantitative contraction rate for HFHR dynamics under a position Poincaré inequality, weighted Hessian and Laplacian bounds, and a compact Sobolev embedding, where the potential function is not necessarily convex. An adapted time-augmented Poincaré inequality yields an explicit rate that improves upon the contraction rate of the underdamped Langevin dynamics. We also give a weak-solution construction and a self-contained spectral proof of the divergence lemma underlying the argument. For HFHR Monte Carlo (HFHRMC) algorithm, which is based on a discretization scheme of HFHR dynamics, we use a path-space Girsanov argument to obtain a non-asymptotic convergence bound and an explicit iteration complexity in total variation distance. The bounds hold for every $α\geq0$ and $γ>0$ and remain regular at the ULD endpoint. Optimizing the iteration complexity bound yields a positive, accuracy-dependent position-diffusion parameter at finite accuracy, while its leading high-accuracy order coincides with that of the optimized ULD endpoint. Our iteration complexity bound improves upon the existing work on HFHR algorithms. Numerical experiments including Bayesian learning problems on real data are provided to illustrate the effect of positive $α$ and its benefit.

Systematic reviews, scoping reviews, mapping studies, and related evidence syntheses are increasingly difficult to conduct with fully manual workflows as search volumes, update cycles, and synthesis requirements continue to expand. At the same time, artificial intelligence, machine learning, and large language models are rapidly entering review practice across query formulation, screening, extraction, categorization, appraisal support, and reporting. Yet the empirical evidence remains uneven, task-dependent, and insufficient to justify unconstrained automation. Existing standards such as PRISMA 2020, PRISMA-S, PRISMA-ScR, PRISMA-P, PRESS, and SWiM remain essential, but none provides an end-to-end operational standard for when AI use is methodologically appropriate, how it should be validated, which review decisions must remain human-led, and how AI involvement should be reported so that readers can audit it. This paper proposes ARISMA, an AI Reporting and Integration standard for Systematic Methods and Analysis. ARISMA treats AI as an inspected, benchmarked, logged, and reversible assistant rather than an autonomous reviewer. It is built around one governing principle: every consequential scientific decision must remain human-interpretable, human-auditable, and human-accountable. The paper contributes a lifecycle taxonomy, process guidance, stepwise recommendations across the review pipeline, a governance and provenance model, a tool-support framework, an AI-integrated reporting checklist, and a validation matrix. It also addresses legal, privacy, infrastructure, and sustainability considerations. The framework was iteratively refined through structured expert consultation. The result is a practical and auditable guideline for responsible AI-assisted evidence synthesis.

Geo-spatial planning, like trip design, is a realistic testbed for LLM agents because it requires grounded tool use, noisy evidence retrieval, and multi-constraint reasoning. Most benchmarks, however, only provide clean geospatial data and tools, missing the open-ended social signals that people use in daily planning. We introduce LifePlanner, a benchmark that enriches map data with large-scale local social media posts and provides access through an MCP toolset. LifePlanner provides an evaluation suite spanning four task categories and three difficulty levels. Experiments show frontier LLMs perform well on simple retrieval but degrade sharply on complex planning, with the Pass Rate dropping to 40.2%. Results show that failures mainly stem from incomplete evidence acquisition from such a large multimodal database, imprecise tool use, and weak constraint integration rather than model size or reasoning length, suggesting that future progress requires effective grounded planning instead of scaling alone.

We introduce grounded glossary generation, a structured task requiring models to recover semantically meaningful Sanskrit phrases and produce translation-grounded meanings from a sloka-translation pair, formalizing the traditional patha commentary practice as an evaluable NLP objective. We construct a benchmark of 31,316 sloka-translation-glossary triples from the Valmiki Ramayana and Srimad Bhagavatam, paired with two metrics: Jaccard for phrase recovery and Meaning Faithfulness for semantic consistency. Across zero-shot, few-shot, and instruction fine-tuned variants of Gemma-3n-E4B, Gemma-3-12B, Phi-4, and Qwen3.5-9B, instruction fine-tuning substantially outperforms prompting, while explicit segmentation yields gains. Error analysis identifies over-segmentation of sandhi and samasa compounds as the dominant failure mode, pointing to morphological modeling as the key bottleneck for faithful Sanskrit lexical decomposition.

Product linking, the entity-resolution task of mapping merchant product records to canonical catalog products, consolidates fragmented listings so downstream search, recommendation, and advertising see one clean entry per product. At marketplace scale, billions of noisy, multi-category records must be resolved against tens of millions of canonical products, where scoring every candidate with a single model is either too weak for the hard cases or too costly for the easy ones. We present a production retrieve-then-match cascade that spends computation in proportion to difficulty: retrieval surfaces plausible matches, a lightweight text cross-encoder auto-resolves the high-confidence majority, and an agentic multimodal vision-language model settles the ambiguous remainder by inspecting product images and issuing web searches for evidence that is in neither record. The cross-encoder is distilled from millions of dual-VLM-consensus labels, retiring human annotation from the training set, and is calibrated to auto-accept links at a 98% precision bar validated against a smaller operator-certified audit. The agent is a self-hosted open-weight model that reaches a closed frontier VLM's precision at a four-point recall cost (88% versus 92%) for roughly one-seventh the per-pair cost, with no fine-tuning. Per-pair cost spans nearly five orders of magnitude from the cheap cross-encoder to the frontier VLM, so escalating only the hard tail to the agent raises end-to-end link coverage from the cheap stage's 68% to 77%.

Standard neural architectures often fail to generalize to longer inputs for dynamic programming (DP) targets. We investigate what makes this hard geometrically. Every finite min-plus DP is a shortest path on a DAG, which is equivalently a tropical polynomial whose extended Newton polyhedron encodes the decision boundary of which path wins. We prove these three descriptions (graph, polynomial, polyhedron) form isomorphic semirings at two levels --- formal polynomials and their computed functions --- connected by operations that characterize all structural redundancies. We then address the length-generalization question geometrically: does the decision boundary at length $T$ decide the boundary at $T+1$? We present two structural negatives. The semiring's two native ways to reduce dimension (setting a variable to each identity) are neither injective nor always closed within the DP. Series and parallel composition fail to construct all DAG topologies from smaller sub-DAGs, and even all terminal-only operations do not capture all DP compositions.

Spoken-language analysis via prompt-based domain-adaptive models is a promising direction for low-resource, non-invasive dementia screening, but such models remain internally opaque. We study the interpretability of the Domain-Adapted models via Prompt-based Fine-tuning (DAPF) framework, which casts dementia detection as diagnosis-related masked-token prediction. We interpret DAPF and strong baselines using a variety of probing and analysis techniques, finding that DAPF achieved the best overall performance (accuracy=0.83 and macro-F1=0.83) with diagnosis most recoverable from its [MASK] representation. However, this representational advantage did not extend to token-level explanation faithfulness. DAPF attributions primarily reflected language task vocabulary, discourse markers, and transcription artifacts, with perturbation tests showing weak or negative effects. This suggests that its masked-token interface determines diagnosis information without producing faithful token-level explanations.

Multimodal product models can complete missing e-commerce attributes, yet current methods still optimize attribute-answer accuracy without verifying visual support, conflate transient prediction with persistent index admission, and lack explicit risk control over factually incorrect or visually unsupported values. We address these gaps with Counterfactual Visual Evidence-Guided Selective Attribute Indexing (CVE-SAI), which first infers and freezes an ontology-constrained candidate from the primary image and attribute question without catalog text, and then decides whether that candidate should enter the index. Focus-Zone Distortion (FZD) constructs an attribute-specific visual-dependence proxy through a controlled counterfactual intervention, and Evidence-Guided Attention Redistribution (EGAR) uses the proxy to refine ontology-constrained scoring. The canonical candidate is frozen before evidence necessity, evidence retention, nuisance-transformation stability, and candidate-specific catalog-text conflict audits; catalog text can only tighten admission and cannot revise the candidate. Independent family-level calibration selects one policy with a simultaneous one-sided finite-sample bound under a 5% unsafe-admission budget. Experiments on five visual attributes derived from Amazon Berkeley Objects show that CVE-SAI improves attribute inference and evidence localization, achieves the highest certified admission coverage under the shared risk protocol, and yields the strongest controlled retrieval performance with the lowest unsafe auto-induced exposure among automatic-admission systems. Separating inference from admission therefore enables visually supported attribute completion to improve retrieval while limiting persistent index contamination.

As people adopt transformer-based language models (e.g., ChatGPT and Gemini) for an increasing number of use-cases, it is important to know how such models learn and represent the meaning of the language, and to be more informed about what language is. This document is an attempt to help the reader understand how linguistic meaning (i.e., semantics) is approached from different fields of scientific and philosophical examination. I also explain three primary semantic theories: formal semantics, grounded semantics, and distributional semantics then compare how transformer-based language models differ from how humans learn language.

Main sequence stars of spectral types late F through M exhibit systematic variability in photometric light curves, particularly when they are young. Rotational modulation of starspots manifests as quasi-sinusoidal variability, which enables the measurement of rotation periods. Variability can also be stochastic, as in stellar flaring. However, since measurements of stochastic processes depend on the time of observation, they are typically noisier. Considering that different manifestations of variability have unique observational nuances, models that naturally unify these are incredibly useful for stellar characterization. Towards this goal, we have developed EncoTESS: a Time Series Foundation Model (TSFM) trained on a subset of TESS 2-min light curves. EncoTESS is specifically designed to handle the observational noise, heteroskedastic measurements, irregular sampling, and large data gaps common to TESS data. It is also ~1% of the size of a typical literature TSFM, so can be run easily on a modern laptop. EncoTESS encodes light curves into a fixed-size latent parameter space, which can be used to infer physical stellar properties and recovers light curve summary statistics well. EncoTESS outperforms rotation period and variability amplitude as age indicators for stars that have not converged onto the slow rotator sequence yet; broadly these include K and M stars less than ~100 Myr, and M stars less than ~1 Gyr. We focus on age inference as an application of EncoTESS in this work, but other downstream tasks such as stellar classification could also be explored. The architecture of EncoTESS enables its future extension to TESS light curves of all cadences, and additional surveys such as Kepler and the upcoming PLATO mission. The core EncoTESS framework and library of encodings produced for the stars used in this work are publicly available at https://github.com/philvanlane/encotess.

A latent world model trains its decoder on latents anchored to observations, then deploys it on the model's own free-running rollout, hundreds of steps past the last observation. Rollout-Decoded Reconstruction (RDR) closes this gap with a single loss term that free-runs the model during training exactly as evaluation will, decodes every rollout latent, and penalizes reconstruction error against ground truth. The term adds no parameters, costs training-time compute only, and reduces to the standard objective at weight zero, so every comparison in this paper is a one-flag A/B. On the chaotic Kuramoto-Sivashinsky equation, RDR raises valid prediction time (the time to first crossing of normalized error 0.5) from $3.87 \pm 0.23$ to $6.97 \pm 0.42$ time units at an identical 193,568 parameters, a $1.80\times$ improvement confirmed on seeds never used in selection and holding in 10 of 10 preregistered configurations at ratios of 1.71-2.50$\times$. The results come from a single system; a sweep in which the advantage grows with latent width is descriptive, and control experiments on two classic tasks are preliminary.

The imperfective paradox provides a useful test of compositional semantic analysis. Recent work constructs an NLI benchmark and reports that models frequently infer completed telic events from progressive descriptions, attributing this behavior to a Teleological Bias. It further argues that prompting interventions cause a Calibration Crisis. We reexamine the benchmark and conclusions and show that it is substantially affected by conceptual and evaluation mis-specifications. We identify three conceptual mis-specifications. In particular, Aspectual Reduction affects the benchmark construction, analysis, experiments, and conclusions. Under a strict NLI standard, 76% of Group A instances do not explicitly rule out culmination. In our native-speaker annotation, 38% of Group A examples and 29% of the Group C examples were judged to permit an alternative interpretation. To control these issues and lexical variation, we construct Lexically Matched Minimal Pairs. At the evaluation level, we formulate event-semantic NLI as a Multi-step Reasoning Problem and assess both intermediate semantic decisions and final predictions. Our results show that models often do not affirm culmination but nevertheless accept the corresponding simple-past hypothesis, a pattern we characterize as Sufficiency Bias. We further show that prompting interventions produce a Decision Shift among labels without reliably improving the underlying semantic understanding and reasoning. Intermediate and oracle-guided analyses identify two additional failure modes: errors in compositional aspectual classification and Surface-form Attraction toward surface-associated answers. Our experiments on Qwen-7B with suitable prompts, GPT-5.4, and Qwen-72B provide initial evidence for the context sensitivity of aspectual classification and suggest that these models can achieve performance comparable to that of human annotators.

Activation steering can be embedded directly into a language model's weights, shaping behaviour without inference-time intervention and offering a way to encode alignment prior to release. However, models are routinely fine-tuned after deployment, and it is unknown whether embedded interventions survive this. We study the stability of embedded steering for refusal suppression and brevity induction across five instruction-tuned models (3B-14B) under non-adversarial SFT and RLHF. Behaviourally, preservation tracks the training data: steering degrades when optimisation pressure contradicts the targeted behaviour and persists otherwise, with refusal ablation losing 64% of its effect on average under SFT. Mechanistically, however, the weight edit survives almost untouched even where behaviour reverts: mean vector recovery is $ρ= 0.004$, and the fine-tuning update along the steering direction is near-orthogonal to its pre-edit weight pattern (mean $\cosθ= 0.074$). When steered behaviour degrades, fine-tuning does not achieve it by dismantling or reversing the steering mechanism itself. Embedded steering is therefore mechanistically durable but functionally vulnerable, and requires behavioural re-validation after downstream training.

Multi-teacher on-policy distillation (MOPD) distills several domain-expert teachers into a single student by minimizing per-domain reverse-KL divergence on the student's own rollouts. Existing approaches typically fix the per-domain data mixture before training, overlooking the fact that different domains converge at substantially different rates: some plateau early while others continue to improve throughout the training budget. A fixed mixture therefore wastes compute on fast-converging domains and undertrains slower-converging ones. To address this, we propose D$^3$-MOPD (Dynamic Domain ScheDuling for MOPD), a zero-overhead scheduler that repurposes the per-domain reverse-KL signal already produced during training to adapt the domain mixture online. Running asynchronously outside the training process, an off-process watcher periodically tracks each domain's KL trajectory, estimates remaining headroom and current improvement rate, and accordingly adjusts the domain sampling ratios without altering the core training loop. Our D$^3$-MOPD scales naturally to arbitrary numbers of domains, and the expected benefit grows as more domains introduce more diverse convergence patterns for the scheduler to exploit. On a Qwen3.6-35B-A3B student distilled from four domain-expert teachers, D$^3$-MOPD closes 97% of the average student-to-teacher performance gap, compared with 63% for vanilla MOPD, reaches the same peak performance with an approximately 3$\times$ reduction in rollout steps, and surpasses the specialist teachers on three of seven benchmarks.

Robust POMDPs (RPOMDPs) generalize classical POMDPs to the setting where exact transition probabilities are not known -- rather, they are only known to belong to some uncertainty set of values. In this work, we study the problem of solving RPOMDPs with general omega-regular objectives, which subsume a broad class of objectives such as reachability, safety, and linear temporal logic (LTL) objectives. We show that, for (s,a)-rectangular RPOMDPs with polytopic uncertainty sets, the problem of solving RPOMDPs under omega-regular objectives can be reduced to solving partially observable stochastic games (POSGs) under omega-regular objectives. Moreover, we show for the first time that reductions can be constructed in both directions, establishing the semantic equivalence between (s,a)-rectangular RPOMDPs with polytopic uncertainty sets and POSGs. This allows us to derive a range of new computational complexity results, including both upper and lower complexity bounds, on solving RPOMDPs with different omega-regular objectives. As a corollary, we also derive new computational complexity results for RMDPs.

Foundation-model post-training usually relies on human labels, preference data, stronger teachers, or executable verifiers. We study Unsupervised Post-Training (UPT): update-bearing adaptation on unlabeled inputs whose learning signal is derived from same-lineage model artifacts rather than an external oracle. We catalog 80 strict UPT methods and organize them by the object that supplies the update signal: a prediction statistic, a sample relation, a self-generated target, or an internal evaluator. Beyond inventory, we show how the choice of internal signal and task structure determines whether post-training improves the model or recursively amplifies error. An orthogonal Input Visibility $\times$ Update Persistence view maps deployment regimes and defines a unified framework for UPT selection and evaluation.

Scientific agents increasingly analyze data, execute code, and produce research artifacts, yet most benchmarks emphasize final answers, isolated programs, or a single domain. We introduce FrontierChallenge, a cross-domain benchmark comprising 300 end-to-end scientific workflows. In this paper, we release and evaluate 97 of these tasks, spanning quantum chemistry, molecular dynamics, materials characterization, analytical chemistry, life science, and electrochemistry/environment. Each task provides fixed inputs and specifies a bundle of required scientific deliverables. We evaluate twelve frontier models with three agent scaffolds. Pass Rate measures the fraction of tasks satisfying the full-completion criterion, while Avg. Score captures partial progress. Each of the best-performing configurations completed only 20 of the 97 released tasks, yielding a Pass Rate of 20.6%. Partial progress translated especially poorly into complete delivery in analytical chemistry and electrochemistry/environment: Avg. Scores reached 87.6 and 94.9, but the highest Pass Rates were only 4% and 0%. Among non-passing Claude Code trajectories, 75.5% still ended with language claiming completion. These findings show that neither high partial scores nor confident claims of completion reliably indicate that a scientific task has been fully delivered, highlighting the need to evaluate end-to-end workflow execution and the completeness of scientific deliverables together.

Multimodal fusion of structural MRI (sMRI) and dynamic functional network connectivity (dFNC) can reveal how brain structure relates to changing functional states. When the same structural latent representation is coupled with multiple states, applying independent vector analysis (IVA) separately to each state can produce unrelated structural decompositions, while forcing identical decompositions may suppress state-specific relationships. In addition, not every subject expresses every dynamic state. We propose masked structural residual IVA (MSR-IVA), a state-aware framework that combines a shared structural representation with state-specific residual adaptations and masks for incomplete state expression. On an Alzheimer's Disease Neuroimaging Initiative cohort, MSR-IVA improved matched source coupling by 6.5% and reduced unmatched dependence by 15.7% relative to the independent pairwise IVA baseline. Among subjects expressing both states, mean absolute cross-state structural source correlation was 0.9177 for MSR-IVA versus 0.2978 for no sharing, demonstrating controlled structural sharing that preserves source correspondence while allowing state-specific adaptation.

Retrieval-Augmented Generation (RAG) enhances large language models by grounding outputs in external knowledge, improving factuality and reducing hallucinations. At the same time, the retrieval-augmented pipeline introduces new robustness and security risks, including corpus poisoning, backdoor attacks, privacy leakage, and fairness violations. Despite rapid progress in this area, existing surveys remain limited in their treatment of attacker objectives, threat models, and stage-specific defenses across the full RAG pipeline. This survey presents a unified and pipeline-aware overview of RAG robustness. We formalize threat models over the corpus, retriever, and generator, and organize attacks into three main objectives: accuracy, privacy, and fairness. We further review defenses from a pipeline-aware perspective, covering the retrieval, rerank, generation, and traceback stages. In addition, we summarize robustness benchmarks and explainability methods for more deeply evaluating and explaining RAG robustness.

Evaluations of generative language models frequently interpret observable behavioral traits, such as political stance, brand inclination, and normative framing, as manifestations of model weights, post-training alignment, or prompting. This interpretation risks conflating a foundation model with the multi-layered production system through which its outputs are ultimately served. Modern inference stacks support runtime interventions capable of modifying generation while model parameters remain frozen. We examine inference-time framing bias: systematic runtime steering of generated text toward institutional, ideological, or commercial frames without requiring changes to the underlying model parameters. We formalize the Inference Attribution Problem and establish an observational non-identifiability result showing that, under black-box observation alone, behaviorally equivalent deployed systems may arise from structurally distinct combinations of model parameters and inference policies. Consequently, observed behavioral bias does not uniquely identify the architectural layer responsible for it. We further characterize Probability Placement as a deployment pattern in which undisclosed commercial influence is embedded within an ostensibly organic assistant response through systematic probability-mass reallocation, distinguishing it from explicit token-auction mechanisms for generative advertising. Finally, we discuss implications for behavioral auditing, inference provenance, confidential computing, cryptographic attestation, the EU AI Act, the Digital Services Act, and advertising-disclosure principles. We argue that governance of generative systems must increasingly distinguish between auditing a model and auditing the deployed system that ultimately speaks.

System-level simulation is an essential tool for exploring the rapidly expanding design space of LLM serving systems, where real deployments remain costly and often infeasible. However, modern LLM serving now evolves faster than human-driven simulator development can track, and emerging workloads and mechanisms, from agentic workflows to disaggregated serving, no longer fit the monolithic simulation pipeline that existing simulators assume. Each new mechanism therefore demands an invasive rewrite, leaving a widening development gap between deployed serving systems and the simulators that model them. To close this gap, we present Simthesizer, a framework that realizes agent-driven simulator development. Simthesizer introduces a composable simulator infrastructure that uniformly expresses the complete serving workflow, including the control decisions that coordinate it, and realizes it as a unified dynamic graph in Simthesizer simulator. Synthesizer agent, a harnessed coding agent, then lowers natural-language feature requests onto this abstraction under simulator-specific guardrails and fidelity validation, evolving one shared simulator instead of building a new one for every feature. Under the same coding agent and harnesses, extensions built on Simthesizer follow a vLLM-based real system with 2.51% average throughput error, versus 6.03% for extensions built on existing simulators. On identical workloads, Simthesizer also simulates up to 284.96x and 23.19x faster than two state-of-the-art simulators, LLMServingSim2.0 and Vidur, respectively.

AI-based structured interference rejection has grown more popular because deep learning approaches can outperform traditional methods by jointly considering the signal of interest (SOI) and the signal mixture (SOI plus interference). This work builds on a previous AI-enabled approach utilizing autoregressive transformer-based models by adding a Finite Scalar Quantization (FSQ) tokenizer layer which aims to improve the interference rejection performance while keeping overall latency to a minimum. Additionally, we experiment with other inference optimization techniques with the goal of speeding up inference without much accuracy loss. We explore this space with an experiment where the SOI is a digitally modulated radio frequency (RF) signal and the structured interference is a digital television signal, an extremely common type of Orthogonal Frequency-Division Multiplexing (OFDM) transmission. Our results achieve low latency and increased interference rejection over traditional techniques and prior work with other AI-enabled methods. We demonstrate the benefits of the AI-enabled approaches via audio metrics such as Perceptual Evaluation of Speech Quality (PESQ). Additionally, we explore a variety of applications and detail how our interference rejection algorithm may be used in operationally-relevant scenarios.

Large Language Model (LLM) agents increasingly solve long-horizon tasks through multi-turn interactions with users and external tools. In these settings, relevant task information often unfolds over time rather than being fully specified at the initial prompt. Service agents make this challenge especially concrete: users may clarify or revise their goals, while tool responses provide information needed for subsequent decisions. Thus, a final reward alone cannot indicate which actions contributed to resolving the task. Recent methods rely on comparative evidence from other trajectories or resampled continuations, or on separately constructed step-level learning signals, to refine credit. However, a completed rollout already records how information and errors flow between agent actions. We introduce Influence-Aware Policy Optimization (IAPO), which represents each rollout as a typed influence-dependency graph over trainable agent actions, with user and tool observations serving as evidence. IAPO converts support-use and failed-use structure into routing weights that redistribute the same trajectory-level advantage. Experiments with Qwen3-4B and Qwen3-8B demonstrate superior performance over multi-turn reinforcement learning (RL) baselines across three service-agent benchmarks: ${τ^2}$-Bench, UserBench, and AgentChangeBench. BFCL-v4 Multi-Turn further shows that these gains do not compromise multi-turn function-calling performance. This work advances the understanding of credit assignment in multi-turn user interactions and provides a principled approach to training service agents from sparse outcome feedback.

With the rapid development of large-scale pre-trained language models based on Transformer architectures, their high computational and memory costs have become a major obstacle to deployment, especially in resource-constrained environments. Traditional pruning methods typically depend on full gradient-based importance estimation, and they necessitate prior finetuning of the model to achieve satisfactory performance. This process often results in intolerable resource consumption. This paper proposes REP-LIE, a new approach to enable resource-efficient pruning during the process of finetuning. REP-LIE leverages the gradients of LoRA low-rank matrices to estimate the importance of weights without requiring full gradient computation. To address the inherent randomness in importance estimation, a stability score is introduced, serving as the basis for iterative pruning of unimportant model parameters. The pruned model is further finetuned through lightweight updates, eliminating the need for full-parameter optimization in the process of finetuning. Extensive experiments on both medium-scale encoder models and large-scale generative models (LLaMA-7B and Mistral-7B) demonstrate that REP-LIE still achieves competitive performance compared to existing approaches.

We study the Lipschitz stability of attention retrieval in hyperbolic spaces. Existing methods lack deterministic guarantees on attention-weight preservation under finite-precision representations. We introduce HCC+, a theoretical framework exploiting three properties of the Poincaré ball: exponential volume growth enabling query-independent boundary truncation; logarithmic covering radius of hyperbolic 1-centers enabling dimension-independent critical-key identification; and a packing bound with constants independent of the embedding dimension. We prove two deterministic guarantees: for exact retrieval, the per-layer attention deviation is bounded by 10\% of its ideal value; for soft attention, the total variation distance decays as $O(1/\sqrt{n})$, the rate of finite-sample variance. As a consequence of the guarding mechanism, the framework achieves a storage reduction factor of $6.1\times$ relative to FP16. We provide the first deterministic, query-independent retrieval certificate in non-Euclidean geometry.

Physics-Informed Neural Networks (PINNs) have emerged as an important class of numerical methods for solving partial differential equations (PDEs). However, during the late-stage optimization process, further parameter updates often yield diminishing accuracy improvements while increasing computational costs. To address this issue, this paper proposes a Physics-Informed Error Field Learning (PIEFL) framework for PINNs. Unlike conventional approaches that continuously approximate the solution field using a single network, PIEFL introduces an auxiliary error network after the primary network achieves satisfactory accuracy and shifts the learning objective from the solution field to the error field. By deriving error control equations under physical constraints, the error network learns the discrepancy between the current approximation and the exact solution, and the learned error correction is combined with the primary prediction to improve solution accuracy. The proposed framework avoids continuous optimization of the entire solution space and focuses computational resources on correcting existing prediction errors. Moreover, PIEFL requires no modification to the primary network architecture, making it compatible with existing PINN models and applicable as a general post-training optimization strategy. Numerical experiments on representative PDEs demonstrate that PIEFL achieves higher solution accuracy under the same computational budget, validating its effectiveness in improving the performance of PINNs.

Continual knowledge graph embedding updates entity and relation representations as a graph grows. Existing methods primarily address catastrophic forgetting, but entity admission also changes the candidate universe of every compatible query. A historical answer can therefore lose rank even when its score and its ordering among old entities are preserved. We formalize this effect as candidate-set interference and introduce Matched Excess-Outranker Regularization (MEOR), a host-level objective that compares smooth answer-relative newcomer pressure with score-blind, structurally matched old references. Its one-sided penalty acts only when newcomer competition exceeds the matched reference, preserving the host learner's signal for legitimate new entities. Across eight paired runs on ENTITY-ComplEx, MEOR improves historical current-universe mean reciprocal rank (MRR) by 0.0057 over replay and reduces candidate-set interference by 0.0055, with one-sided 95% lower bounds of 0.0052 and 0.0051, respectively. It satisfies the preservation criteria for old-universe ranking and newcomer acquisition and improves historical current-universe MRR over persistent calibration, matched maximum regularizer (MMR), and unmatched old regularizer (UOR). Direct ablations support each component of its reference construction and aggregation. Adding MEOR also improves historical ranking in all ten reported FBInc-S and FBInc-L host and backbone settings, with every paired 95% confidence interval excluding zero. These results establish candidate admission as a distinct source of continual rank loss and show that it can be controlled without replacing the underlying embedding architecture or continual learner.