Papers

Multi-view multi-label feature selection aims to identify a compact and informative feature subset from heterogeneous views while preserving discriminative information for multiple labels. Existing methods are generally developed from specific modeling perspectives and incorporate mechanisms tailored to particular data characteristics. Designing suitable feature selection algorithms across datasets with diverse and heterogeneous characteristics still relies heavily on expert knowledge and substantial manual effort, imposing considerable time and labor costs that severely hinder the practical adoption of feature selection. To address this problem, we propose FeatureHospital, a Skill-driven multi-agent framework for automated multi-view multi-label feature selection algorithm design. FeatureHospital first diagnoses the target dataset to identify its feature selection issues. Based on the diagnosis, specialist agents equipped with domain Skills then prescribe corresponding optimization strategies and Loss terms for different issues. After that, the resulting prescriptions are reconciled to remove overlaps and resolve conflicts before being integrated into a compact dataset-specific objective. Finally, the constructed objective is optimized to select the final feature subset. Experimental results demonstrate that FeatureHospital can construct effective feature selection algorithms for different datasets based on their individual characteristics.

Class-imbalance handling is routinely evaluated on a single benchmark dataset, and the resulting conclusions are reported as if they were properties of the method. We show this practice is unsafe. On the public Kaggle credit-card fraud dataset, under a leakage-free nested cross-validation protocol in which the decision threshold is selected on a held-out inner validation fold, a plain Random Forest at the default 0.5 threshold attains F1 = 0.861 +/- 0.021, and threshold tuning yields it no benefit (delta-F1 = -0.002). Read alone, this supports an appealing conclusion: for a well-calibrated ensemble, imbalance handling is unnecessary. We then apply the identical protocol to 45 binary tasks spanning imbalance ratios from 1:1.5 to 1:178 (2,025 model fits, four model families). The conclusion reverses. Random Forest benefits most from threshold tuning across the suite (delta-F1 = +0.101 +/- 0.134), not least, while three other families replicate their fraud-dataset behaviour almost exactly. SMOTE likewise harms the fraud dataset but helps across the suite (mean delta-F1 = +0.076; 138 wins, 39 losses; Wilcoxon p = 2.7e-17). Two further results. Threshold-tuning benefit is non-monotonic in the imbalance ratio: near zero below 1:5, peaking at +0.120 in the 1:15-1:40 band, declining to +0.045 beyond 1:100 - explaining why the fraud dataset, at 1:577, is an unrepresentative place to study the question. And we reject an intuitive heuristic: validation-set calibration error does not predict tuning benefit (expected calibration error r = -0.087; Brier r = +0.137), so calibration diagnostics cannot tell a practitioner whether tuning is worthwhile. We release the protocol, the 45-task harness, and all per-run metrics.

UAV on-board vision systems are widely used for different activities, including monitoring in no-fly zones. In this case, the vision-equipped UAV streams a video to a ground server where an operator assists its activities. The latency of video transmission has a profound impact on the effectiveness of the operator assistance. However, most techniques available for video transmission still incur significant latency costs. In this paper, we propose a graph convolutional neural network-assisted (GCN-Assisted A2C) deep reinforcement learning (DRL) system model to find the optimal pixel-correlated area of a suspicious object. We combine the Lagrangian dual form with gradient descent to prevent lack of convergence and over- and under-penalization constraint violation during latency optimization. The proposed system model sends a sub-group pixel-correlated area of the frame from the UAV to the server rather than the transmission of the whole video frame. The proposed framework utilizes the GCN model to explore hidden representations of feature-correlated groups of pixels. Moreover, the GCN supervises the A2C model, which selects a subgroup to enhance transmission latency, thus supervising the training of UAV actions in A2C. Experimental results show that GCN-assisted A2C reduces video frame transmission latency together with false detection rate in UAV vision systems over other DRL and state-of-the-art models.

Adolescent Idiopathic Scoliosis (AIS) is a prevalent spinal deformity in adolescents that, if left untreated, can result in severe health outcomes. Traditional screening methods are limited by subjective interpretation, reliance on professional expertise and low scalability. To address these challenges, we present ScoliGait dataset, which comprises 1,516 gait video clips paired with corresponding X-ray records. We also introduce TokenSTFormer, a novel model that tokenizes spatial and temporal semantics to enhance feature representation and convergence. Our model achieves state-of-the-art performance, surpassing vanilla Vision Transformer encoder across key metrics, including accuracy of 0.79. This study highlights the potential of leveraging holistic motion features derived from gait video and attention-based models for scalable, cost-effective AIS screening, paving the way for future clinical applications in scoliosis detection.

How should we assess whether large language models can perform mathematical invention? I argue that this question is currently underspecified: mathematical creativity is not one capacity but several mechanistically distinct modes of meaning-making - reflexive introspection on mathematical practice, analogical import from the sciences, problem-driven construction, and the bridging of distant domains - together with a further, cross-cutting distinction between meaning pursued because a pattern was observed and meaning pursued because it is strategically wanted, a distinction I develop through the case of conjecture-formation. These mechanisms are likely non-substitutable, so that competence in one does not transfer to the others. Grounding each in a historical case study and in an architecture-level account of current transformer-based systems, I suggest that today's models concentrate their competence in modes shaped by recombination and search over existing building blocks; if that description holds, the remaining modes are out of reach in principle, not just slower - though whether it holds is itself the open, empirical part. Because proof is getting cheaper as AI improves at generating it - a shift the field's own leading voices are now diagnosing - mathematical value is migrating toward the modes current systems cannot yet perform, and evaluations of AI mathematical ability should be organized around this taxonomy rather than around aggregate benchmarks that conflate it.

Retrosynthesis is a cornerstone of drug discovery and organic synthesis. While data-driven deep learning models have shown remarkable progress, they autonomously learn reaction patterns from extensive datasets with limited integration of established chemical knowledge as priors. To address this limitation, we introduce RetroMPA, a molecular property-aware, post-hoc enhancement module that injects chemical knowledge into the retrosynthesis pipeline. Rather than functioning as an independent SMILES sequence generator, RetroMPA is a broadly applicable, model-agnostic chemical filter designed to recalibrate and optimize the predictive pathways of existing algorithms. This plug-and-play framework integrates seamlessly with a range of data-driven retrosynthesis methods, enhancing outputs without modifying model architecture or requiring resource-intensive retraining. By leveraging a property-aware latent embedding space, RetroMPA consistently improves top-1 accuracy across eight representative retrosynthesis models by an average of 5.50% on USPTO-50K. Furthermore, we validate its scalability on the large-scale USPTO-Full dataset, achieving an average improvement of about 2.03% across both template-based and template-free architectures. Wet-lab experiments provide preliminary support for the practical utility of the framework. These syntheses confirmed viable, previously unreported substrate combinations for classic reaction paradigms—specifically, Suzuki-Miyaura coupling, Bucherer reaction, and Friedel-Crafts acylation—suggesting that RetroMPA can operate beyond mere data fitting. The code is open-sourced at https://github.com/MengzhouLu/RetroMPA.

Multivariate time-series forecasting faces a structural dilemma: sharing one temporal predictor across variables is parameter-efficient but forces heterogeneous variables through an identical history-to-future map, whereas learning an independent predictor per variable restores flexibility at a cost that grows with the product of variable count, context length, and horizon. We argue that this dilemma dissolves once the object being compressed is the forecasting operator rather than the observed series. Auditing per-variable linear history-to-future maps across standard benchmarks, we find that a phase-locked seasonal component paired with a compact residual operator outperforms a dense phase-blind reference in most audited settings. The residual transport is also directional: lag-invariant alternatives consistently underperform asymmetric history-to-future maps. Guided by this structure, we propose AsyTO, an Asymmetric Temporal Operator that factorizes the tensor of per-variable operators into shared but distinct history-reading and future-writing temporal modes with per-variable mode-wise gains, complemented by a low-rank periodic prototype and a cycle-separable factorization of the temporal modes. Each forecast reads only its own variable's history, so parameters and compute grow linearly in the number of variables. Across eleven benchmarks and multiple forecast horizons, AsyTO attains the best lightweight error in 30 of 44 dataset-horizon settings, locating at the accuracy-compute Pareto frontier.

Accurate protein structure prediction is fundamental to structural biology because protein structure underlies molecular function and provides a basis for mechanistic interpretation. Recent advances in deep learning have transformed the field from multiple sequence alignment (MSA)-driven monomer folding into broader frameworks capable of modeling protein complexes and increasingly heterogeneous molecular systems. Existing reviews have summarized this progress from the perspectives of representative models, application domains, and protein design. Building on these efforts, this review focuses on the methodological evolution of the field itself. It examines recent developments through three closely related dimensions: representations and data, architectures and learning strategies, and confidence and evaluation. Within this perspective, the field is organized into four methodological phases and three cross-cutting transitions: from explicit evolutionary coupling features and early contact prediction to learned sequence representations in AlphaFold2, RoseTTAFold, and ESMFold; from protein-only monomer folding to increasingly integrated modeling of heterogeneous molecular systems in AlphaFold-Multimer, RoseTTAFoldNA, and AlphaFold3; and, more recently, from prediction-oriented structure inference to design-oriented generative modeling in RFdiffusion and related frameworks. This framework provides a clearer understanding of how methodological shifts have shaped the capabilities, limitations, and practical roles of recent models.

Capability development is routinely inferred from behavioural thresholds, from final checkpoints, or from what a decoder can read out of a hidden state. These quantities need not identify the same event. We study a 30M-parameter recurrent-depth relational reasoner in a closed, oracle-defined world, using dense behavioural trajectories, two training surfaces, preregistered pre-arrival hidden-state probes, prospectively checked evaluability, and explicit untrained and negative controls, holding the training-time and inference-time axes separate throughout. Behaviour first: under one frozen acquisition criterion, three-hop competence cost 70 logical epochs on the symbolic surface and 13,055 on the verbal surface, a 186.5-fold contrast, after which verbal four-hop competence cleared in 8 logical epochs. Across the 13,055-epoch grind, four-hop held-out behaviour never exceeded 3/40 and ended at 0/40. Internal measurement next: on the verbal surface a linear probe recovered future-answer identity before behavioural arrival at 0.056159 against uniform chance 0.025, an untrained control of 0.024758 and a population frequency baseline of 0.048309 (p = 0.012987; 16/40 answer classes contributing). Analogous pre-arrival accessibility survived the surface change, reaching 0.1020 against a zero-step control of 0.0460 (p = 0.000999) at the upstream structural position and 0.0618 at the readout comparator (p = 0.004), with 21/40 classes contributing. Finally, the natural attempt to track that accessibility across training was not cleanly evaluable: probe eligibility is defined by behavioural arrival, so the measured population changes with the measurand. Behavioural competence, internal accessibility, and training-time development are distinct observables, and neither behaviour nor decoder accessibility identifies the computation training acquired; causal intervention is the necessary next step.

Neural autoregressive models have rapidly emerged as powerful emulators of high-dimensional chaotic systems, yet their long-term instability and error growth remain poorly understood, leading to ad-hoc solutions. Here, we develop an eigenanalysis framework that reveals the dynamical origin of this error growth. By analyzing the Jacobian of the learned one-step update map with respect to the state, we show how inference-time error growth, and thus model stability, is governed by its spectral radius. Direct-step architectures (models that predict the next state from the previous one) generically admit unstable eigenvalues with magnitudes exceeding one, explaining the rapid divergence of these widely used models. In contrast, integration-constrained models (where the time derivative is estimated and integrated with a higher-order integrator) collapse their eigenspectrum onto the unit circle, yielding neutral stability and a universal linear error-scaling law. The largest eigenvalue of this Jacobian provides an architecture-agnostic, a priori diagnostic of short-term skill, long-term stability, and spectral bias, without requiring an expensive rollout. Leveraging this theory, we introduce a stability-promoting loss that explicitly regularizes Jacobian-driven error amplification, improving both forecast accuracy and dynamical robustness. Demonstrated across $29$ models spanning two architectures, several explicit and implicit integrators, and multiple loss functions on the Kuramoto-Sivashinsky system, our results establish a theoretical foundation for the design and evaluation of neural emulators of chaotic multi-scale dynamics. More broadly, our framework is a step toward the kind of a priori stability analysis that numerical analysis provides for discretizations of differential equations and that scientific machine learning currently lacks.

Reinforcement learning (RL) with group-relative advantages has become the de facto standard for post-training language model reasoners. However, when optimizing multiple reward objectives, existing methods typically scalarize the reward vector with a fixed weighted sum before group-wise standardization. We show that this design leads to two fundamental problems: rollouts with distinct reward profiles can receive identical advantages, and all objectives are optimized with fixed relative weights regardless of their current level of saturation. Consequently, an objective whose rewards are already near their upper bound can retain substantial influence when its rewards still vary within rollout groups. We introduce Saturation Aware Advantage Reweighting for Multi-Reward Policy Optimization (SA-MRPO), which standardizes each reward objective independently and adaptively discounts its contribution according to a batch-level estimate of objective saturation. This changes the relative contribution of each objective according to its observed reward headroom. We further derive an exact condition under which saturation aware reweighting reverses the sign of a rollout's aggregate advantage. Across mathematical reasoning with two- and three-objective reward combinations, SA-MRPO improves the harder correctness objective over GDPO in 12 of 15 benchmark comparisons, with gains of up to $5\%$ on AIME24. On adaptive reasoning it improves accuracy on all five benchmarks, by $3.8\%$ on average and up to $9.2 \%$ on AMC23, and on coding benchmarks it improves pass rate by up to $2.3\%$, while in all settings maintaining the easier objectives near their already satisfied levels. Additional experiments characterize the accompanying reward tradeoffs and sensitivity to corrupted rewards.

Reliable global ocean forecasting is critical for climate monitoring, marine navigation, and extreme event early warning. Physics-based ocean forecasting models impose prohibitive computational costs, while existing deep learning approaches predominantly rely on structured-grid architectures, incurring unnecessary computation on masked land cells and enforcing uniform resolution across dynamically heterogeneous ocean regions regardless of local flow complexity. Here we present OceanLight, an efficient global ocean forecasting framework innovatively combining geometry-adaptive unstructured mesh tokenization with a graph neural network (GNN) backbone. OceanLight achieves pointwise forecast accuracy and kinetic energy spectral fidelity exceeding both operational numerical analyses and state-of-the-art AI-based models, while surpassing all AI-based ocean models in geostrophic balance consistency. Furthermore, OceanLight demonstrates reliable mesoscale eddy representation, capturing coherent ocean structures beyond pointwise statistical optimization. These capabilities are delivered with a 62% reduction in GPU memory consumption and 70% reduction in FLOPs relative to structured-grid baselines. Our unstructured mesh representation establishes a generalizable paradigm for scalable data-driven oceanography.

Large language models (LLMs) are increasingly deployed as agents that rely on system prompts to use tools and complete tasks. Such deployments impose distinct operational requirements, including appropriate tool use, concise prompts and solution paths, and compliance with safety and formatting policies. For many practitioners, however, assembling domain-specific supervised data to post-train models to meet these requirements is infeasible. We introduce CAPO (Constraint-Aware Prompt Optimization), a primal-dual method that combines pool-based rewrites with adaptive constraint weighting to optimize system prompts under explicit operational constraints. Across agentic benchmarks, CAPO more reliably reaches empirically feasible operating points while improving task performance. CAPO also generalizes beyond agentic settings, achieving strong results on assistant-style evaluations with output-format and safety/privacy constraints. We further introduce DCAPO (Dynamically Trained CAPO), which trains a feedback- and dual-conditioned rewriter with pool-based GRPO while keeping the task agent frozen. Across task agents of different sizes, DCAPO produces a feasible prompt in every evaluated domain and matches or improves the task accuracy achieved by the evaluated baselines. A surrogate analysis characterizes how finite-pool and discrete-rewrite errors enter the inexact primal-dual procedure.

Existing agent benchmarks ask whether the agent finished the task. We ask whether it finished it within policy. We introduce Fiducia-bench, a benchmark for the governability of financial agents—whether they escalate when obligated, abstain when required, and leave an auditable trail—and use it to study a question no prior benchmark addresses: does decomposing an agent into components degrade its governance? It does, and the mechanism is specific. Policy-relevant facts discovered by one component are attenuated at the handoff boundary before reaching the component that must act on them. In a 626-episode experiment across 100 KYC/AML task variants, two models, and three architectures, a 32B open-weights model attenuated 0% of discovered facts under a single-loop baseline, 56% under a fixed pipeline, and 85% under an orchestrator-subagent architecture (all at constraint distance 2). A stronger model (gpt-4.1-mini) attenuated 3-6% under the same conditions, suggesting the governance cost of decomposition is partly a function of model capability. Critically, the same mechanism produces both under-escalation and over-escalation, depending on whether the dropped fact was a risk signal or an exculpating one. The benchmark, all tasks, and the verification harness are open-source

LLM-based data-analysis tools are increasingly used to help users analyze messy spreadsheets and workbooks, from answering questions over uploaded files to generating code, summaries, and visualizations. These systems are often evaluated by the correctness of their final downstream answers. However, reliable data analysis also depends on an earlier step: understanding what the dataset contains before solving the requested task. For complex workbooks, this Data Exploration step includes identifying the logical tables behind physical sheets, interpreting column semantics, recovering keys and relationships, and detecting quality issues. In current tools and benchmarks, this step is usually left implicit, creating a gap between downstream task performance and the dataset understanding needed for reliable, human-checkable analysis. Our key contribution is to identify this overlooked gap, make Data Exploration a first-class evaluation target, and show through downstream experiments that stronger Data Exploration support improves task performance. To evaluate dataset understanding directly, we introduce two benchmark settings: a real multi-sheet workbook benchmark based on a Vitamin D study dataset, and an extension of DSBench with schema-fixed Data Exploration artifacts. In both settings, systems are evaluated by the quality of a structured artifact capturing tables, columns, semantic roles, relationships, and profiling signals. Our results show that strong LLMs and data-analysis agents still miss important logical structure even when they read spreadsheet content. Furthermore, explicit Data Exploration support often improves downstream correctness, suggesting it should be treated as a first-class, inspectable stage in LLM data-analysis workflows and a natural human-in-the-loop checkpoint where domain experts can review and correct the artifact before downstream analysis proceeds.

Brain parcellation and classification are typically evaluated in isolation, yet downstream AD detection performance depends on their interaction. We decouple these components and systematically benchmark fast deep learning parcellation methods (SynthSeg+, OpenMAP-T1) against the FreeSurfer (FS-HV) clinical baseline through down- stream AD classification on OASIS-1. Our factorial design evaluates three parcellation methods, two volumetry strategies (hard vs. soft), and four classifier paradigms (clinical thresholds, supervised feedforward networks, ensemble methods, and foundation models with zero/few-shot prompting), with all results quantified using BCa Bootstrap 95% confidence intervals.

Post-hoc Graph Neural Network (GNN) explainers commonly follow a Perturb-Query paradigm, inferring the importance of graph elements based on queried predictions to perturbed inputs. However, such perturbations often introduce substantial distribution shift, undermining the reliability of the queried predictions used to derive explanations. While existing efforts mainly improve perturbed graphs or stabilize model predictions on them, we revisit the perturbation mechanism itself. We show that the widely used Element-wise Masking(EM) suppresses edge-induced messages toward zero, causing deterministic scale contraction that accumulates across message-passing layers, a phenomenon we term Scale Drift. Consequently, prediction changes under EM may conflate information corruption with deviations in propagation scale. As a scale-stable alternative to EM, we introduce Noise Corruption (NC), which perturbs each message through matched-norm random-direction corruption while preserving the expected squared message norm. Building on NC, we propose NICE, a Noise Corruption-based explanation framework, which learns a Stochastic Restoration Boundary (SRB) under NC-induced uncertainty, balancing target-prediction restoration against compactness. Furthermore, Boundary-Integrated Gradient (BIG) converts this boundary into edge attributions by accumulating each edge's contribution to reducing restoration risk along the restoration path. Experiments across multiple benchmarks demonstrate stronger explanation performance and model faithfulness while confirming that NC substantially reduces the Scale Drift induced by masking.

Graph anomaly detection aims to identify nodes that deviate from normal behavioral patterns within graphs. However, existing methods largely rely on the homophily assumption, which makes it difficult to distinguish spurious affinities and to capture the diverse behaviors of normal nodes,limiting their robustness in complex real-world scenarios. To address this problem, we propose RagGAD, an unsupervised graph anomaly detection framework based on rationale-aware conditional Gaussian mixture normalizing flow. RagGAD introduces an adaptive rationale disentangler to disentangle stable rationales from spurious correlations within node interrelationships, and further decomposes stable rationales into robust and fragile components. The learned rationales capture underlying interaction patterns that characterize normal behaviors under varying conditions, while anomalies emerge as deviations associated with unstable or spurious correlations. To model the intricate distributions of normal and abnormal nodes, RagGAD integrates rationale-non-rationale Gaussian mixture modeling with a robust-fragile rationale mixture learning strategy. By mitigating spurious homophilic correlations and embracing the heterogeneity of normal patterns, RagGAD identifies anomalies as low-density regions within a structure-aware distribution space. Extensive experiments on multiple benchmark datasets demonstrate that RagGAD outperforms state-of-the-art methods.

Generative Artificial Intelligence (GenAI) can produce high-quality essays, code, and design artefacts, challenging the validity of conventional assessments that rely on single-point submissions and product-only grading. This paper proposes a design framework called "Dynamic Evidence Collection Ecosystem" that shifts assessment toward continuous, authentic, multi-source evidence of student learning over time. The framework collects process evidence through iterative artefacts, design logs, activity rounds, self-reflection, and peer collaboration, supported by an AI-enabled layer for learning analytics, formative feedback, and transparency. The approach is grounded in recent assessment-redesign scholarship in AI-rich contexts and aligned with contemporary views of authenticity in assessment. This paper builds on the hypothesis that academic integrity is strengthened when it is treated as an assessment design rather than as an AI detection problem. The tools have limitations and risks of use that carry academic penalties. This paper presents an implementation scenario to support institutional adoption.

Automated checking pipelines increasingly place one language model as the checker and another (or the same one) as the fixer. We ask whether that wiring changes what the checker reports. Measuring false alarms on human-verified-correct ProcessBench traces with the present task held byte-identical, we find that a completed audit -> repair episode already in the model's context lowers false alarms in 15 of 15 model x wording combinations, by 2.8 to 11.5 percentage points against a length-matched non-audit control, a 9 to 25% reduction relative to that control. The direction contradicts what the accumulated-message literature predicts: an episode whose audit reported an error lowers false alarms further still, at all five wordings on the model where that manipulation lands cleanly, though a negativity asymmetry predicts more flagging. Decomposing the episode finds repair content and audit verdict complementary: different components carry the effect on different model families. Signal-detection analysis locates the change in the threshold rather than in discrimination – the criterion moves in 15 of 15 combinations and survives correction in 13 while d' survives in none, though the d' test is half as sensitive by construction – and a hand audit of 50 false alarms finds 82% simply wrong, so at this operating point the shift need not be harmful. With reasoning enabled the effect keeps its relative size on both models tested, and the threshold reading holds there too.

Reliable uncertainty quantification (UQ) is essential for deploying large language model (LLM) agents in complex interactive environments. Existing UQ methods largely rely on local signals, such as token probabilities, predictive entropy, or per-step confidence, and therefore overlook the long-range dependencies through which errors accumulate across an execution trajectory. As a result, they may fail to identify agent failures whose causes originate several reasoning or interaction steps before the final answer. We propose RUPA (Relational Uncertainty Propagation for Agents), a trajectory-level UQ framework for LLM agents. RUPA represents an execution history as a directed trajectory graph in which reasoning states, tool interactions, and environment feedback are nodes connected by temporal and semantic dependency edges. It then propagates uncertainty over this graph to capture how execution risk accumulates and transfers across interaction steps. The propagated signal is combined with trajectory-level behavioral features and goal-alignment information to produce a confidence estimate for the full agent trajectory. We evaluate RUPA on representative agent benchmarks, including $\tau$-2, Terminal-Bench-2, and GAIA, using 6 open-source LLMs spanning multiple model families. Experimental results show that RUPA consistently outperforms existing UQ methods by providing more accurate uncertainty estimates, enabling earlier failure detection, and improving uncertainty-guided agent execution across diverse agent tasks. These results demonstrate that explicitly modeling relational dependency is crucial to reliable UQ for long-horizon LLM agents, providing a practical foundation for trustworthy agent execution.

Automatic emotion assessment can benefit from combining neural and behavioral signals, but many multimodal approaches rely on separate, modality-specific feature-extraction pipelines before fusion. This paper presents MUPA\textsuperscript{2}E, a unified perception framework that processes facial video and electroencephalography (EEG) through a single shared asymmetric-attention backbone. Facial video is represented through axis-folded frame tokens, while EEG is processed either as a raw multichannel waveform or projected into the spatial domain for multimodal fusion. The framework is evaluated on the DMER dataset under a stratified subject-independent protocol, comparing unimodal video, unimodal EEG, and fused video–EEG configurations with per-channel and merged EEG projections. Using the original recordings, with shorter trials zero-padded to match the longest duration, merged fusion at stride~$30$ achieves the highest validation performance and a test accuracy of $70.07\%$. Further analysis revealed that recording duration is unevenly distributed across the affective classes, making the padding pattern a potential classification cue. Controlling for this factor by cropping all recordings to a common duration of $20$ seconds yielded a test accuracy of $62.71\%$, providing a stricter duration-controlled assessment of the framework in which differences in recording length are removed as a potential classification cue. These findings demonstrate the feasibility of processing structurally different neural and visual signals within a compact unified architecture while highlighting the importance of controlling duration-related cues in affective datasets.

Large language models produce outputs presented as discoveries - new proofs, conjectures, or molecules. Whether such an output that appears creative is truly original and effective is hard to establish: open-ended outputs require subjective judgment, the output may replicate something seen in training, or the task may be too simple to need creativity. We present ALPS (Austin-Law Proof-Synthesis), a benchmark that designs a task to measure valid creativity: producing a solution that is original and can be proven correct. Each instance is a single equational law, certified to require either the construction of an infinite mathematical structure satisfying the law, or a proof that no such structure exists. Submissions are verified by automated proof checking with no human involvement, and a public generator produces new instances without limit, so LLMs are never evaluated on problems they may have seen. A portfolio of eight configurations of leading automated provers resolves 2.2% of the 4,141-law evaluation pool, and a twentyfold budget increase adds 0.6%: the obstacle is not compute, but the absence of any method that produces the tailored structure each law requires. Under a fixed protocol, the strongest reasoning model we test succeeds in 14% of instances on the proof side, but none on the construction side. The remaining 97.2% of the pool is unresolved at every configuration and budget we test. We release ALPS in full: the corpus, the generator, and the automated judge.

Labeling large text corpora with LLM teachers has become a practical route to training data at scale. At millions of items, hand-labeling every batch is not feasible, and two questions dominate: what label quality a teacher buys per dollar, and how to keep a fleet of GPU workers busy under skewed, failure-prone workloads. We present a simple, reproducible pipeline that addresses both. First, a work-stealing ring pool: each worker owns a queue, drains it first, and then steals from ring successors, with exactly-once task claims via atomic conditional writes and crash tolerance via stale-claim sweeping. The claim protocol requires only a compare-and-set primitive from its storage layer; we implement it on a single SQLite file, which makes the reference implementation dependency-free and the experiments reproducible on one machine. Second, a memory-aware concurrency rule that sizes per-node parallelism by how many model copies fit on the GPU, so the same code runs safely across device sizes. Third, a relabeling benchmark methodology in which the teacher relabels a public dataset that already has gold labels, so quality reduces to an agreement measurement and cost follows from measured throughput. Under skewed load the pool sustains up to 3.4 times the throughput of static sharding while matching it at zero skew, loses 0 of 2,000 tasks when half the workers are killed mid-run (static sharding loses 953), and yields measured quality and cost points for an instruction-tuned teacher on irony and sentiment tasks. All experiments run on public data and commodity hardware; code, tests, and run logs are released.

Large Reasoning Models (LRMs) rely on explicit chain-of-thought (CoT) reasoning and large context windows to achieve strong performance on complex tasks, but these features also introduce new attack surfaces. We show that LRMs' reasoning processes can be systematically steered by prepending counter-aligned few-shot conversations containing explicit CoT traces, leading to unsafe generations on harmful queries and unwarranted refusals on benign ones. We formalize this attack as SRCF (Steering Reasoning via Counter-Aligned Few-shot Conversations) that operates solely through a flexible conversational interface and requires no access to the model's parameters and gradients. Our key insight is that SRCF exploits an adversarial generalization issue that induces a representation drift, causing the representations of benign and harmful inputs to shift in a similar direction. This observation motivates our post-training defense, ARCF (Aligning Reasoning via Counter-Aligned Few-Shot Conversations), which exposes models to counter-aligned conversational contexts while enforcing aligned targets. ARCF is compatible with existing post-training methods and consistently improves safety and helpfulness without degrading utility.

We study how instruction-tuned LLMs arbitrate direct conflicts between system and user instructions. We introduce a benchmark of 41 paired constraints with deterministic verifiers and evaluate eight models under matched baseline, conflict, and same-channel control conditions. Behaviourally, the models split into three regimes by System Authority Delta: hierarchy-respecting models use the system channel as an authority signal, anti-hierarchy models follow the system less often than their same-channel baseline predicts, and no-effect models show little channel sensitivity. Llama-3.1-8B is the strongest anti-hierarchy case in our suite, following the system in only 0.10 of conflict trials. We use this behavioural failure case to ask whether user-preferring arbitration reflects the absence of an internal conflictresolution signal. It does not: on Llama-3.1-8B, the conflict outcome is linearly decodable from residual-stream activations at 0.97 balanced accuracy, 17 percentage points above a metadata-only baseline, with analogous signals on Qwen2.5-7B and gpt-oss-20b. Steering with a layer-12 mean of four per-conflict logistic-regression directions raises genuine system compliance from 0.132 to 0.530, while directions selected mainly for pooled separability steer poorly. User-preferring conflict resolution can therefore coexist with a readable internal arbitration signal, and successful intervention depends on the geometry of the readout rather than probe accuracy alone

Target-only post-training can improve performance in a specialized domain while degrading behaviors that a general-purpose base model acquired before adaptation. We study this problem when target-domain data are available but a representative replay corpus is not. We propose self-specialized teacher distillation (SSTD), a two-stage procedure that first trains a copy of the base model into a domain teacher, then distills its token distribution to a student on prefixes sampled from the student itself. Teacher training combines standard target supervision with base-aware key-token weighting and distribution alignment to the frozen base model; on-policy distillation then places domain feedback on states the student can encounter at inference time. On financial numerical reasoning, medical question answering, and legal holding identification, SSTD retains much of the target improvement of direct fine-tuning while improving the mean score on the evaluated general suite by 4.8–5.0 points at the reported operating point. The pattern persists across Qwen3 sizes and on Gemma backbones. SSTD requires neither an external teacher nor general replay data.

Video generation is central to AI-powered content creation. Aligning generated videos with human preferences is a key criterion for evaluating generation quality. Despite significant progress in visual quality, three key challenges remain. First, the reliability of reward signals is constrained by the quality of human preference data, which is often affected by subjective noise and bias. Second, standard scalar reward models collapse multi-aspect human preferences into a single value, leading to the loss of dynamic trade-offs across multiple preference dimensions. Third, in policy optimization, the widely adopted KL divergence imposes primarily local constraints and may fail to capture the global structure of human preferences. To address these challenges, we propose a unified preference-aware learning framework for video generation. First, we introduce elite-guided filtering to calibrate preference data and construct reliable supervision for reward model training. We then model video quality as a multidimensional reward distribution to capture the uncertainty inherent in human preferences, and use the Wasserstein distance to align the learned reward distribution with the empirical human preference distribution. Finally, we introduce Wasserstein-based distributional alignment into GRPO, guiding policy optimization to better match the global structure of human preferences over videos. Experiments on reward modeling and video generation demonstrate that our approach improves the reliability of reward signals and the perceptual consistency of generated videos. Our code is available at https://github.com/alignhs26/ahs.

Agentic security uses large-language-model (LLM) agents to plan, dispatch, and interpret security tools. As these systems move from demonstrations to deployed products, practitioners repeatedly encounter the same operational failures. We systematize these failures through a hands-on evaluation of ten widely used static, dynamic, cloud, orchestration, and AI red-teaming tools for unattended pipelines. We introduce a four-dimensional Integration Friction Index that separates one-time engineering cost from recurring organisational, legal, and maintenance cost. We then derive quantitative regularities that explain recurring failure modes. Modelling an agentic security system as stochastic LLM policies wrapped by a deterministic mediator, we show that long-lived sessions lose resident evidence with phase count, while short-lived sub-agents extend the usable horizon according to the compression ratio between raw evidence and its summary. We show that a two-stage verdict cascade multiplies scorer likelihood ratios, but provides little benefit when scorer errors correlate. We show that treating unevaluable outcomes as attack failures biases downstream measurements toward evasive and severe responses. We formulate planner-versus-worker model routing as a knapsack problem and derive a closed-form execution cap for heavy-tailed tools, eta* = alpha v/c. Finally, we show why scope and budget enforcement cannot be delegated to system prompts: prompts do not constrain what actually executes. Inspectra, our implemented platform, serves as a worked instantiation, with mechanisms labelled shipped, partial, or planned, including those that did not work.

Early multi-agent LLM systems often used role-separated teams, yet scaling agent count yields diminishing returns on repository-level coding tasks. Recent alternatives treat agents as passive tools (subagents), yet this removes the benefits of agent interaction entirely. We study whether a subagent paradigm can support a middle ground: minimal agentic cooperation without the overhead of large multi-agent teams. We introduce Adversarial Review (AR), a minimal cooperative code-review protocol in which a main coding agent works with a reviewer and a critic agent. The reviewer evaluates code, while the critic audits the review through structured disagreement before the main agent edits. On LiveCodeBench, AR achieves the highest pass rate among tested methods, outperforming a five-agent baseline while using only three agents. On SWE-PRBench, naive AR exposes a false-consensus failure mode, where agents converge on agreement without sufficient evidence, but a single prompt iteration that adds disagreement explicitly achieves the highest F1 among tested methods. On SWE-bench Verified, AR also shows improvements over the baselines on repository-level coding tasks. Together, AR demonstrates that cooperative code review does not require many agents or complex communication structures: it requires that disagreement be minimal, structured, and evidence-grounded.