Safe navigation for mobile robots demands policies that remain reliable under the high-consequence perception uncertainty of cluttered environments. Yet most existing safe reinforcement learning (RL) methods assess safety through average cumulative cost. Such metrics can mask dangerous tail-risk behaviors. To address this, we propose a framework that trains risk-sensitive policies through Conditional Value-at-Risk (CVaR) constrained optimization on an off-policy TD3 backbone and evaluates their safety margins post-training through neural network reachability verification. During training, the policy is optimized under CVaR constraints on cumulative costs, promoting sensitivity to high-cost tail outcomes rather than average behavior alone. After training, we compute action reachable sets under bounded observation uncertainty using Taylor Model analysis, yielding a safety rate metric that quantifies the proportion of evaluated states at which the policy's reachable action set remains within prescribed safety margins. A key finding is that policies trained with CVaR constraints maintain larger safety margins from obstacles across evaluated states. This makes them significantly more amenable to formal reachability verification. Experiments across ten navigation scenarios and six baselines show that our method achieves a 98.3\% success rate, the highest safety verification rate among all compared methods, while revealing that average cost rankings and reachability-based safety rankings can diverge. This indicates that reachability verification captures risks which are missed by empirical cost metrics alone. We further validate our approach on a physical Clearpath Jackal robot, demonstrating successful sim-to-real transfer.

Channel state information (CSI) provides a widely available sensing modality for human and environment perception, but existing CSI sensing models usually rely on task-specific supervised training and require substantial labeled data for each task, device, user, or environment. This limits their scalability in practical deployments where unlabeled CSI is abundant but labeled data is costly to collect. In this paper, we present CSI-JEPA, a self-supervised predictive representation learning framework for label-efficient, multi-task Wi-Fi sensing. CSI-JEPA learns reusable temporal-spectral representations from unlabeled CSI samples by predicting latent features of masked channel regions from visible context. To better match the physical structure of CSI, CSI-JEPA tokenizes channel-response amplitude windows along the time and subcarrier dimensions. It then introduces a channel variation-aware masking strategy that samples predictive targets from regions with stronger local temporal and subcarrier-domain variations. After pretraining, the encoder is frozen and used as a backbone, with lightweight task-specific adapters added for downstream sensing tasks. We evaluate CSI-JEPA on seven real-world Wi-Fi sensing tasks spanning diverse objectives and deployment settings. The results show that CSI-JEPA improves downstream sensing performance over competitive baselines, achieving up to 10.64 percentage points mean accuracy gain over state-of-the-art supervised Transformer and matched-budget label savings of up to 98.0%.

Memory systems are critical for role-playing agents (RPAs) to maintain long-horizon consistency. However, existing RPA memory methods (e.g., profiling) mainly rely on recurrent summarization, whose compression inevitably discards important details. To address this issue, we propose a search-based memory framework called BOOKMARKS, which actively initializes, maintains, and updates task-relevant pieces of bookmarks for the current task (e.g., character acting). A bookmark is structured as the answer to a question at a specific point in the storyline. For each current task, BOOKMARKS selects reusable existing bookmarks or initializes new ones (at storyline beginning) with useful questions. These bookmarks are then synchronized to the current story point, with their answers updated accordingly, so they can be efficiently reused in future grounding rounds. Compared with recurrent summarization, BOOKMARKS offers (1) active grounding for capturing task-specific details and (2) passive updating to avoid unnecessary computation. In implementation, BOOKMARKS supports concept, behavior, and state searches, each powered by an efficient synchronization method. BOOKMARKS significantly outperforms RPA memory baselines on 85 characters from 16 artifacts, demonstrating the effectiveness of search-based memory for RPAs.

Learning of continuous exponential family distributions with unbounded support remains an important area of research for both theory and applications in high-dimensional statistics. In recent years, score matching has become a widely used method for learning exponential families with continuous variables due to its computational ease when compared against maximum likelihood estimation. However, theoretical understanding of the statistical properties of score matching is still lacking. In this work, we provide a non-asymptotic sample complexity analysis for learning the structure of exponential families of polynomials with score matching. The derived sample bounds show a polynomial dependence on the model dimension. These bounds are the first of its kind, as all prior work has shown only asymptotic bounds on the sample complexity.

Every AI benchmark operationalizes theoretical assumptions about the capability it claims to assess. When assumptions function as unexamined commitments, benchmarks stabilize the dominant paradigm by narrowing what counts as progress. Over time, narrow evaluation reorganizes capability concepts: architectures and definitions are selected for benchmark legibility until evaluation ceases to track an independent object and instead produces a version of the target defined by its own operational assumptions. The result is a trap: evaluation frameworks treat self-reinforcing assessments as valid, both creating and obscuring structural limits on what the current paradigm can accomplish. We introduce Epistematics, a methodology for deriving evaluation criteria directly from technical capability claims and auditing whether proposed benchmarks can discriminate the claimed capability from proxy behaviors. The contribution is meta-evaluative: an audit procedure, a failure mode taxonomy, and benchmark-design criteria for evaluating capability-evaluation coherence. We demonstrate the procedure through a worked audit of Dupoux et al. (2026), a proposal that revises the dominant paradigm's theoretical assumptions at the architectural level while reproducing them in its evaluation criteria, thereby entrenching the constraint it seeks to overcome in a form the evaluation cannot detect.

The primary way to establish and compare competencies in foundation and generative AI models has shifted from peer-reviewed literature to press releases and company blog posts, where model builders highlight results on selected benchmarks. These artifacts now largely define the state of the art for researchers and the public. Despite their prominence, which benchmarks model builders choose to highlight, and what they communicate through this selection, is underexamined. To investigate, we introduce and open-source Benchmarking-Cultures-25, a dataset of 231 benchmarks highlighted across 139 model releases in 2025 from 11 major AI builders, alongside an interactive tool to explore the data. Our analysis reveals a fragmented evaluation landscape with limited cross-model comparability: 63.2% of highlighted benchmarks are used by a single builder, and 38.5% appear in just one release. Few achieve widespread use (e.g., GPQA Diamond, LiveCodeBench, AIME 2025). Moreover, benchmarks are attributed different competencies by different builders, depending on their narrative. To disentangle these conflicting presentations, we develop a unified taxonomy mapping diverging terminology to a shared framework of measured signals based on what benchmark authors claim to measure. "General knowledge application" is the second most popular, yet vaguely defined, category. Qualitative analysis shows many such benchmarks deemphasize construct validity, instead framing results as indicators of progress toward AGI. Their authors claim to measure knowledge or reasoning broadly, yet mostly evaluate STEM subjects (especially math). We argue that highlighted benchmarks function less as standardized measurement tools and more as flexible narrative devices prioritizing market positioning over scientific evaluation. Data: https://hf.co/datasets/matybohacek/benchmarking-cultures-25; tool: https://bench-cultures.net.

Can a committee of weak reasoning-model calls reach the performance of much stronger models? We study verifier-backed committee search as inference-time boosting for reasoning language models. The mechanism is not simply that ``more agents help'': samples expose latent correct solutions, while critics and comparators must recover them without access to the hidden verifier. We formalize this view by separating proposal coverage, local identifiability, progress, and diversity. We prove that coverage can be amplified by repeated sampling, but cannot by itself create useful critics or comparators; reliable amplification requires an additional local soundness signal, such as execution, proof checking, type checking, tests, or constraint solving. We give rank-based bounds showing when local selection errors compose into reliable trajectories, and characterize the proposer-side ceiling: oracle best-of-\(k\) converges only to the mass of task slices on which the proposal system assigns nonzero useful probability. Empirically, on SWE-bench Verified, a single \texttt{GPT-5.4 nano} proposal solves \(67.0\%\) of tasks. Using the same nano model, our critic--comparator orchestration reaches \(76.4\%\) with \(k=8\) proposals, matching the standalone performance of \texttt{Gemini 3 Pro} and \texttt{Claude Opus 4.5} Thinking and approaching the \(79.0\%\) oracle best-of-\(8\) upper bound. Thus, many correct patches are already present in weak-model proposal pools; the main challenge is selecting them. The remaining failures are mostly proposal-coverage failures, indicating shared blind spots that stronger selection alone cannot close.

While generative models have shown promise in pediatric sleep analysis, the latent structure of their multimodal embeddings remains poorly understood. This work investigates session-wide diagnostic information contained in the sequences of 30-second pediatric PSG epochs embedded by a multimodal masked autoencoder. We test whether augmenting embeddings with PHATE-derived per-epoch coordinates and whole-night movement descriptors, persistent homology summaries of the embedding cloud, and EHR yields task-relevant signals. Simple linear and MLP models, chosen for interpretability rather than state-of-the-art performance, show that geometric, topological, and clinical features each provide complementary gains. For binary predictions, feature importance is task-dependent, and more expressive late-fusion models generally perform better, with AUPRC improving from 0.26 to 0.34 for desaturation, 0.31 to 0.48 for EEG arousal, 0.09 to 0.22 for hypopnea, and 0.05 to 0.14 for apnea. We also report Brier score and Expected Calibration Error, where the full fusion model yields the best calibration across all four binary tasks. Our study reveals that latent geometry/topology and EHR offer complementary, interpretable signals beyond embeddings, improving calibration and robustness under extreme imbalance.

Safety evaluations for large language models (LLMs) increasingly target high-stakes National Security and Public Safety (NSPS) risks, yet multilingual safety is typically assessed through translation-only benchmarks that preserve the underlying scenario, and empirical evidence of how language and geopolitical context interact remains limited to a narrow set of language pairs. We introduce \emph{ROK-FORTRESS} https://huggingface.co/datasets/ScaleAI/ROK-FORTRESS_public, a bilingual, culturally adversarial NSPS benchmark that uses the English--Korean language pair and U.S.--ROK geopolitical axis as a case study, separating the effects of language and geopolitical grounding via a \emph{transcreation matrix}: adversarial intents are evaluated under controlled combinations of (i) English versus Korean language and (ii) U.S.\ versus Korean entities, institutions, and operational details. Each adversarial prompt is paired with a dual-use benign counterpart to quantify over-refusal. Model responses are then scored using calibrated LLM-as-a-judge panels, applying our expert-crafted, prompt-specific binary rubrics. Across a dual-track set of frontier and Korean-optimized models, we find a consistent suppression effect in Korean variants and substantial model-to-model variation in how geopolitical grounding interacts with language. In many models, Korean grounding mitigates the Korean language-driven suppression -- with no model showing significant amplification in the other direction -- indicating that, at least in the English--Korean case, safety behavior is shaped by language-as-risk signals and context interactions that translation-only evaluations miss. The transcreation matrix methodology is designed to generalize to other language--culture pairs.

Objective: The primary goal of this study was to systematically examine the impact of commonly used imbalance handling methods (IHMs) on predictive performance in biomedical binary classification, considering the interplay between model complexity and diverse data modalities. Material and Methods: We evaluated five representative IHMs: random undersampling (RUS), random oversampling (ROS), SMOTE, re-weighting (RW), and direct F1-score optimization (DMO), against a raw training (RAW) baseline. The evaluation encompassed three public biomedical datasets: MIMIC-III (tabular), ADE-Corpus-V2 (text), and MURA (image), spanning three common biomedical data modalities. To assess varying model complexity, we employed a range of architectures, from classical logistic regression and random forest to deep neural networks, including multilayer perceptron (MLP), BiLSTM, BERT, DenseNet, and DINOv2. Results: For simpler models such as logistic regression on tabular data, IHMs yielded no significant advantage over the RAW baseline, aligning with prior findings. However, clear benefits were observed for more complex models and unstructured data: (a) ROS and RW consistently enhanced the performance of powerful models; (b) direct F1-score optimization demonstrated utility primarily for unstructured text and image data; and (c) RUS and SMOTE consistently degraded performance and are therefore not recommended. Conclusion: The effectiveness of IHMs depends on both model complexity and data modality. Performance gains are most pronounced when leveraging appropriate IHMs, such as ROS, RW, and DMO, on high-complexity models.

bde is a user-friendly Python package for Bayesian Deep Ensembles with a particular focus on tabular data. Built on an efficient JAX implementation of the sampling-based inference method Microcanonical Langevin Ensembles (MILE), it provides scikit-learn compatible estimators for fast training, efficient Markov Chain Monte Carlo sampling, and uncertainty quantification in both regression and classification tasks.

The field of deep visual recognition is undergoing a paradigm shift toward universal representations. The Platonic Representation Hypothesis suggests that diverse architectures trained on massive datasets are converging toward a shared, "ideal" latent space. This again raises a critical question: is a "Good Embedding All You Need?" In this paper, we leverage this convergence to demonstrate that off-the-shelf embeddings are inherently "good enough" for complex tasks, rendering intensive task-specific fine-tuning unnecessary. We explore this hypothesis within the few-shot learning framework, proposing a straightforward, non-parametric pipeline that entirely bypasses backpropagation. By utilizing a k-Nearest Neighbor classifier on frozen DINOv2-L features, we conduct a layer-wise characterization to identify an optimal feature extraction. We further demonstrate that manifold refinement via PCA and ICA provides a beneficial regularizing effect. Our results across four major benchmarks demonstrate that our approach consistently surpasses sophisticated meta-learning algorithms, achieving state-of-the-art performance.

We study learning when the learned object is executable solver code rather than a predictor. In this setting, correctness is not enough: two solvers may both return valid solutions on the deployment distribution while differing substantially in runtime. Given samples from an unknown task distribution, the learner returns code evaluated on fresh instances by both solution quality and execution time. Our central abstraction is a \emph{solver hint}: reusable structure inferred from samples and compiled into specialized solver code. We prove that the empirically fastest sample-consistent solver from a fixed library generalizes in both correctness and runtime, and that statistically identifiable hints can be recovered and compiled from polynomially many samples. Empirically, we instantiate the framework with LLM code agents on \(21\) structured combinatorial-optimization target distributions across seven problem classes. The synthesized solvers reach mean normalized quality \(0.971\), improve by \(+0.224\) over the average heuristic pool and by \(+0.098\) over the highest-quality heuristic, and are \(336.9\times\), \(342.8\times\), and \(16.1\times\) faster than the quality-best heuristic, Gurobi, and the selected time-limited exact backend, respectively. On released PACE 2025 Dominating Set private instances, the synthesized solver is valid on all \(100\) graphs and runs about two orders of magnitude faster than top competition solvers, with a moderate quality gap. Inspection shows that many gains come from changing the computational scale: replacing ambient exponential search or general-purpose optimization with compiled distribution-specific computation.

Recent text-to-video diffusion transformers generate visually compelling frames, yet still struggle with temporal coherence, often producing flickering, drifting, or unstable motion. We show that these failures leave a clear imprint inside the model: incoherent videos consistently exhibit irregular, fragmented temporal diagonals in their intermediate self-attention maps, whereas stable motion corresponds to smooth, band-diagonal patterns. Building on this observation, we introduce TeDiO, a training-free, inference-time method that reinforces temporal consistency by regularizing these internal attention patterns. TeDiO estimates diagonal smoothness, identifies unstable regions, and performs lightweight latent updates that promote coherent frame-to-frame dynamics, without modifying model weights or using external motion supervision. Across multiple video diffusion models (e.g., Wan2.1, CogVideoX), TeDiO delivers markedly smoother motion while preserving per-frame visual quality, offering an efficient plug-and-play approach to improving dynamic realism in modern video generation systems.

We present PanoPlane, an approach for high-fidelity sparse-view indoor novel view synthesis that reconstructs closed room geometry via panoramic scene completion. Unlike perspective-based methods that generate training views from limited fields of view, PanoPlane leverages $360^{\circ}$ panoramic completion to condition the generative process on the full spatial layout. We propose Layout Anchored Attention Steering, a training-free mechanism that steers attention within the diffusion model's internal representation toward scene's detected planar surfaces at inference time. By directing each unobserved region's attention toward geometrically consistent observed content, our method replaces unconstrained hallucination with grounded surface extrapolation. The resulting panoramic completions provide supervision for 3D Gaussian Splatting, enabling accurate novel-view synthesis across unobserved regions from as few as three input views. Experiments on Replica, ScanNet++, and Matterport3D demonstrate state-of-the-art novel view synthesis quality across 3, 6, and 9 input views, achieving up to $+17.8\%$ improvement in PSNR over the current state-of-the-art baseline without any training or fine-tuning of the diffusion model.

Fast Healthcare Interoperability Resources (FHIR) is the dominant standard for interoperable exchange of healthcare data. In FHIR, electronic health records form a directed graph of resources. Answering clinically meaningful questions over FHIR requires agents to perform multi-step reasoning, filtering, and aggregation across multiple resource types. Prior work shows that even tool-augmented LLM agents (retrieval, code execution, multi-turn planning) often select the wrong resources or violate traversal constraints. We study this problem in the context of FHIR-AgentBench, a benchmark for realistic question answering over real-world hospital data, and frame reasoning on FHIR as a sequential decision-making problem over a queryable structured graph. We implement a multi-turn CodeAct agent and post-train it with reinforcement learning using a custom harness and tools. A LLM Judge provides execution-grounded rewards. Compared to prompt-based, closed-model baselines, RL post-training improves performance while enforcing data-integrity constraints. Empirically, our approach improves answer correctness from 50% (o4-mini) to 77% on FHIR-AgentBench using a smaller and cheaper Qwen3-8B model. We present an end-to-end post-training pipeline (environment building, harness construction, model training and custom evaluation) that reliably improves multi-turn reasoning over structured clinical graphs.

Geospatial foundation models compress multispectral observations into dense embeddings increasingly used in natural-language environmental reasoning systems. A single planetary-scale model, e.g. Google AlphaEarth, handles broad characterization well but may compromise on specialized hydrologic signals. Such generalist models are also often inaccessible, expensive, and require large-scale compute. We propose Mini-JEPAs: a fleet of small sensor-specialized Joint Embedding Predictive Architecture (JEPA) foundation models consulted by a routing agent for specialized questions. We pretrained five 22M-parameter Mini-JEPAs sharing an identical Vision Transformer backbone, JEPA recipe, and 64-d output space, using Sentinel-2 optical, Sentinel-1 SAR, MODIS thermal, multi-temporal Sentinel-2 phenology, and a topography-soil stack. Each Mini-JEPA reconstructs the variable matched to its sensor, with cross-validated $R^2$ reaching 0.97 for elevation, 0.97 for temperature, and 0.81 for precipitation. The five manifolds differ in geometric structure, with global participation ratios from 8.9 to 20.2 and local intrinsic dimensionalities from 2.3 to 9.0. Joint topography-soil and phenology models add predictive value beyond AlphaEarth alone for soil moisture, aridity, and precipitation ($ΔR^2$ up to 0.031). A router LLM reads per-modality references and selects appropriate sensors with a perfect hit rate over a curated question set. In paired LLM-as-Judge evaluation, dual retrieval over AlphaEarth and the routed fleet outperforms AlphaEarth alone on physics-matched questions (Cohen's $d = 1.10$, $p = 0.031$). Locally-trained Mini-JEPAs can be operationalized for hydrologic intelligence with modest compute.

An AI system for professional floor plan design must precisely control room dimensions and areas while respecting the desired connectivity between rooms and maintaining functional and aesthetic quality. Existing generative approaches focus primarily on respecting the requested connectivity between rooms, but do not support generating floor plans that respect numerical constraints. We introduce a text-based floor plan generation approach that fine-tunes a large language model (LLM) on real plans and then applies reinforcement learning with verifiable rewards (RLVR) to improve adherence to topological and numerical constraints while discouraging invalid or overlapping outputs. Furthermore, we design a set of constraint adherence metrics to systematically measure how generated floor plans align with user-defined constraints. Our model generates floor plans that satisfy user-defined connectivity and numerical constraints and outperforms existing methods on Realism, Compatibility, and Diversity metrics. Across all tasks, our approach achieves at least a 94% relative reduction in Compatibility compared with existing methods. Our results demonstrate that LLMs can effectively handle constraints in this setting, suggesting broader applications for text-based generative modeling.

Biomedical retrieval-augmented large language models (LLMs) often face evidence that is incomplete, misleading, or internally contradictory, yet evaluation usually emphasizes answer accuracy under helpful context rather than reliability under conflict. Using HealthContradict, we evaluate six open-weight LLMs under five controlled evidence conditions: no retrieved context, correct-only context, incorrect-only context, and two mixed conditions containing both correct and contradictory documents in opposite orders. In this conflicting-evidence order contrast, where the same two documents are both present and only their order is reversed, accuracy drops for every model and 11.4%--25.2% of predictions flip. To support abstention in these difficult cases, we also evaluate a conflict-aware abstention score that combines model confidence with a detector of evidence conflict. In the two hardest conditions, this score improves selective accuracy over confidence-only, with mean gains of 7.2--33.4 points in incorrect-only (`IC') and 3.6--14.4 points in incorrect-first conflicting (`ICC') conditions across 75%, 50%, and 25% coverage. These results show that conflicting biomedical evidence is both an uncertainty and robustness problem and motivate evaluation and abstention methods that explicitly account for evidence disagreement.

Hospital pharmacists make high-stakes decisions to mitigate drug shortages under uncertainty, time pressure, and patient risk. Interviews revealed that pharmacists focus attention on a small subset of drugs, limiting cognitive effort to the most urgent cases. Motivated by these findings, we formalize a bounded-rational, attention-guided decision framework that dynamically decomposes drugs into a subset for high-cost reasoning and a complementary subset for low-cost monitoring. We develop two agents: an Expert Agent that applies attention weights derived from pharmacist interviews, and a Learner Agent that adapts attention allocation over time through experience. Across simulated scenarios spanning short to long horizons, we show that attention-guided planning supports stable decision-making without complete state reasoning. These results suggest that a primary decision is not what action to take, but where to allocate cognitive effort, and that attention-guided, satisficing strategies can reduce problem complexity while maintaining stable performance.

Vision Transformers (ViTs) enable strong multi-view 3D detection but are limited by high inference latency from dense token and query processing across multiple views and large 3D regions. Existing sparsity methods, designed mainly for 2D vision, prune or merge image tokens but do not extend to full-model sparsity or address 3D object queries. We introduce SToRe3D, a relevance-aligned sparsity framework that jointly selects 2D image tokens and 3D object queries while storing filtered features for reactivation. Mutual 2D-3D relevance heads allocate compute to driving-critical content and preserve other embeddings. Evaluated on nuScenes and our new nuScenes-Relevance benchmark, SToRe3D achieves up to 3x faster inference with marginal accuracy loss, establishing real-time large-scale ViT-based 3D detection while maintaining accuracy on planning-critical agents.

Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness, yet rural regions often lack the specialists and infrastructure needed for early detection. Although cloud-based deep learning systems offer high accuracy, they face significant challenges in these settings due to high latency, limited bandwidth, and high data transmission costs. To address these challenges, we propose a two-tier edge-cloud cascade on the public APTOS 2019 Blindness Detection dataset. Tier 1 runs a lightweight MobileNetV3-small model on a local clinic device to perform a binary triage between Referable DR (Classes 2-4) and Non-referable DR (Classes 0-1). Tier 2 runs a RETFoundDINOv2 model in the cloud for ordinal severity grading, but only on the subset of images flagged as referable by Tier 1. On a stratified APTOS test split of 733 images, Tier 1 reaches 98.99% sensitivity and 84.37% specificity at a validation-tuned high-sensitivity threshold. The default cascade forwards 49.52% of test images to Tier 2, reducing cloud calls by 50.48% relative to using a cloud-based model for all images. In the deployed 4-class output space (Class 0-1 / Class 2 / Class 3 / Class 4), the cascade obtains 80.49% accuracy and 0.8167 quadratic weighted kappa; the cloud-only baseline obtains 80.76% accuracy and 0.8184 quadratic weighted kappa. On APTOS, the cascade cuts cloud use by about half with a modest drop in grading performance. Index Terms: Diabetic Retinopathy, Edge-Cloud Cascade, MobileNetV3-small, RETFound-DINOv2, Retinal Screening, tele-ophthalmology

We investigate whether behavior cloning is sufficient to produce active perception in a structured object-finding task. A low-cost robot arm equipped with a wrist-mounted egocentric RGB camera must reposition to center a partially visible plant before triggering a grasp signal, requiring actions that improve future observations. The model predicts joint commands directly from low-resolution RGB images under closed-loop control. We show that low-resolution egocentric vision is sufficient for reliable task completion and that predicting relative joint deltas substantially outperforms absolute joint position prediction in our setting. These results demonstrate that visually grounded active perception can emerge from behavior cloning in a reproducible setting.

Inferring spatially resolved gene expression from histology images offers a cost-effective complement to spatial transcriptomics (ST). However, existing methods reduce this task to a simple morphology-to-expression mapping, where visual similarity does not guarantee molecular consistency. Meanwhile, single-cell data has amassed rich resources far surpassing the scale of ST data, yet it remains underexplored in vision-omics modeling. Furthermore, current approaches commit to a monolithic paradigm with bottlenecks, unable to balance expressive flexibility with biological fidelity. To bridge these gaps, we propose DUET, a novel dual-paradigm framework that synergizes parametric prediction and memory-based retrieval under cellular inductive priors. DUET implements a parallel regression-retrieval paradigm, adaptively reconciling the outputs of its complementary pathways. To mitigate aleatoric vision ambiguity, we incorporate large-scale single-cell references to impose molecular states as biological constraints for faithful learning. Building upon structural refinement, we further design a lightweight adapter to dynamically assign branch preference across spatial contexts to achieve optimal performance. Extensive experiments on three public datasets across varied gene scales demonstrate that DUET achieves SOTA performance, with consistent gains contributed by each proposed component. Code is available at https://github.com/Junchao-Zhu/DUET

In recent years, a variety of powerful LLM-based agentic systems have been applied to automate complex tasks through task orchestration. However, existing orchestration methods still face key challenges, including strategy collapse under reward maximization, high gradient variance with opaque credit assignment, and unguided skill evolution whose decisions are typically made by directly prompting an LLM to judge rather than derived from principled training signals. To address these challenges, we propose SkillFlow, a flow-based framework that takes a trainable Supervisor as the agent and a structured environment with dynamic skill library and frozen executor, automating task orchestration through multi-turn interaction. SkillFlow employs Tempered Trajectory Balance (TTB), a regression-based flow-matching loss that samples trajectories proportional to reward, preserving diverse orchestration strategies rather than collapsing to a single mode. The same flow objective yields a jointly learned backward policy that provides transparent per-step credit assignment at zero additional inference cost. Building on these flow diagnostics, a recursive skill evolution mechanism determines when to evolve, what skills to create or prune, and where decision gaps lie -- closing the loop from training signal to autonomous capability growth. Experimental results on 14 datasets show that SkillFlow significantly outperforms baselines across question answering, mathematical reasoning, code generation, and real-world interactive decision making tasks. Our code is available at https://anonymous.4open.science/r/SkillFlow-E850.

Large language models (LLMs) have revolutionized natural language processing. Understanding their internal mechanisms is crucial for developing more interpretable and optimized architectures. Mechanistic interpretability has led to the development of various methods for assessing layer relevance, with cosine similarity being a widely used tool in the field. On this work, we demonstrate that cosine similarity is a poor proxy for the actual performance degradation caused by layer removal. Our theoretical analysis shows that a layer can exhibit an arbitrarily low cosine similarity score while still being crucial to the model's performance. On the other hand, empirical evidence from a range of LLMs confirms that the correlation between cosine similarity and actual performance degradation is often weak or moderate, leading to misleading interpretations of a transformer's internal mechanisms. We propose a more robust metric for assessing layer relevance: the actual drop in model accuracy resulting from the removal of a layer. Even though it is a computationally costly metric, this approach offers a more accurate picture of layer importance, allowing for more informed pruning strategies and lightweight models. Our findings have significant implications for the development of interpretable LLMs and highlight the need to move beyond cosine similarity in assessing layer relevance.

Fairness in toxicity classification involves three integrated axes: ranking, calibration, and abstention. Training-time interventions and post-hoc safety mechanisms cannot be evaluated independently because the former determines the efficacy of the latter. We compare Empirical Risk Minimization (ERM), instance-level reweighting, and Group DRO across these axes, combined with temperature scaling, confidence-based abstention, and per-identity threshold optimization. Evaluation uses subgroup AUC, BPSN/BNSP AUC, error gaps, and per-subgroup Expected Calibration Error (ECE) with bootstrap CIs ($n = 1000$). We report four findings. (1) Calibration disparity is a hidden fairness violation. ERM has near-perfect aggregate calibration ($0.013$) but is significantly miscalibrated across all identity subgroups ($+0.029$ to $+0.134$). (2) Training interventions reshape rather than eliminate disparity. Reweighted ERM improves ranking (BPSN AUC $+0.06$ to $+0.12$) but worsens the calibration-fairness gap by up to $+0.232$. Group DRO eliminates calibration disparity but only by becoming uniformly miscalibrated globally (ECE $0.118$). (3) Post-hoc methods inherit training failure modes. Temperature scaling fails because miscalibration is non-uniform. Confidence-based abstention works under ERM but breaks under DRO, where the risk-coverage curve rises with deferral. (4) Abstention itself is unfair. Confidence-based deferral helps background content far more than identity-mentioning content. We argue that SRAI fairness requires a multi-axis framework: methods that differ only in aggregate ranking can differ sharply in failure modes that determine real-world harm.

Deep neural networks have achieved strong performance in genomic sequence classification; however, relating their predictions to biologically meaningful sequence patterns remains challenging. In this work, we present AttnGen, an attention-guided training framework that embeds interpretability directly into the optimization process. AttnGen computes nucleotide-level importance scores using an attention mechanism and progressively suppresses low-contribution positions during training. This encourages the model to focus its predictions on a compact set of informative regions while reducing reliance on noisy sequence elements. We evaluate AttnGen on the standardized demo_human_or_worm benchmark, a binary classification task over 200-nucleotide sequences. With moderate masking, AttnGen achieves a validation accuracy of 96.73%, outperforming a conventional CNN baseline with 95.83% accuracy, while also exhibiting faster convergence and improved training stability. To assess whether the learned importance scores reflect functionally relevant signal, we conduct perturbation-based analysis by removing high-saliency nucleotides. This causes accuracy to drop from 96.9% to near chance level on a 3,000-sequence evaluation set, indicating that the model relies on a relatively small subset of informative positions. Our analysis shows that masking 10--20% of positions provides the most favorable trade-off between predictive performance and interpretability. These results suggest that attention-guided masking not only improves classification performance but also reshapes how models distribute importance across sequence positions. Although this study focuses on short genomic sequences, the proposed approach may extend to more complex interpretable sequence modeling settings.

Distilling reasoning traces from strong large language models into smaller ones is a promising route to improve intelligence in resource-constrained settings. Existing approaches face a fundamental trade-off: offline distillation from teacher-generated traces provides high-quality, sample-efficient supervision but suffers from distributional drift: during training, the student model conditions on teacher-generated prefixes, whereas during inference the student autoregresses on self-generated prefixes, leading to compounding errors over long reasoning trajectories. Meanwhile, on-policy or self-distillation methods better match the student's inference-time distribution, but require costly online sampling and often produce low-quality traces in early training. We propose a principled offline reasoning distillation framework that preserves the efficiency and supervision quality of offline teacher-generated data while correcting teacher-student distribution drift. It adaptively emphasizes teacher supervision that is better aligned with the student's on-policy distribution. Evaluations on mathematical reasoning benchmarks of GSM8K, MATH, MATH500, and harder held-out competition-style tasks, including AMC, AIME, and OlympiadBench, show that our method improves reasoning accuracy over prior offline distillation algorithms and yields more stable reasoning traces while preserving instruction-following capabilities. Our work shows that lightweight, distribution-correction-aware training can substantially strengthen offline reasoning distillation without online rollouts.

Irregularly sampled multivariate event streams remain a stubbornly difficult modality for generative modeling: tokenization-based approaches break down when inter-event intervals vary by orders of magnitude, and neural temporal point processes are bottlenecked by window-level numerical quadrature. We (i) propose SurF, a generative model that uses the Time Rescaling Theorem (TRT) as a learnable bijection between event sequences and i.i.d.\ unit-rate exponential noise, enabling a single model to be trained across heterogeneous event-stream datasets; (ii) three efficient parameterizations of the cumulative intensity that scale to long sequences; and (iii) a Transformer-based encoder for multi-dataset pretraining. On six real-world benchmarks, SurF achieves the best reported time RMSE on Earthquake, Retweet, and Taobao, and is within trial-level noise of the strongest specialist on the remaining three. Under a strict leave-one-out protocol, the held-out checkpoint beats every classical and neural-autoregressive baseline on 5/6 datasets and beats every baseline on Amazon and Earthquake, an initial step toward foundation models over asynchronous event streams.