New Machine Learning (cs.LG) papers, Tuesday 20 October 2026
512 papers announced on arXiv. Get them ranked for you, free →
Dependency, Compression, and Synergy: A Unified Information-Theoretic View of Multimodal Learning
Recent advances in multimodal foundation models have intensified the need to understand how different modalities share, preserve, and complement information. Mutual Information (MI), the Information Bottleneck (IB), and Partial Information Decomposition (PID) provide…
Web Agent Agentic Reinforcement Learning Decision Model Under Multi-Cost and Failure Risk
AI summaryProposes a constrained agentic RL model that unifies page access, search, and API calls with cost and CVaR risk constraints, reducing cost‑per‑success by 22‑31% and failures by 18‑26% while boosting completion rates under budget limits.
PolyGraph: Plan-Buffered CUDA Graphs for Adaptive Speculative Decoding
AI summaryPolyGraph enables adaptive speculative decoding by storing the current batch‑draft shape in GPU memory and reusing a single CUDA Graph; it matches baseline latency while cutting pre‑serve build time 42× and reduces padding waste up to 18.7%.
Forecasting Patient Trajectories from the Electronic Health Records with Action-Conditioned World Models
AI summaryPredicts patient trajectories by learning action‑conditioned latent state transitions in EHR data, enabling multi‑step rollouts and action perturbations. The model retains predictive accuracy comparable to larger baselines with a 30× smaller parameter count.
Measuring the Influence of Modality-Specific Node Representations on Graphs
Multimodal attributed graphs combine graph structure with node text and image representations in recommendation and multimedia analysis. Graph predictors propagate these representations across connected nodes, so one input can change predictions at its own and nearby nodes.…
SAGE-CP: When Marginal Coverage Hides Route-Conditional Failures in Adaptive VLM Ensembles
AI summarySAGE-CP trains a fixed routing policy for VLM ensembles and calibrates each exit path separately with Mondrian split conformal prediction, guaranteeing finite‑sample coverage per route. It raises worst‑route coverage from 0.623 to 0.887 while cutting calls, latency, and energy by ~60%.
Composable Parametric Demographic Injection for Values Survey Simulation
AI summaryPARADE injects demographic shifts into LLMs via reusable LoRA modules for each attribute, separating task knowledge from demographic effects. In simulations of World Values Survey and SocioBench, it reduces population KL divergence by up to 52.8% while preserving predictive accuracy.
DUET: Decoupled Upper- and Lower-Body Execution for Task-Agnostic Humanoid Loco-Manipulation
Changing a humanoid robot's manipulation policy should not require retraining the controller that maintains the robot's balance. This study introduces DUET, a decoupled execution framework that enables locomotion and manipulation controllers to operate simultaneously by…
Language-Forcing: Advancing Humanoid Loco-Manipulation via Semantic Alignment of Whole-Body World Action Models
World Action Models (WAMs) jointly predict robot actions and future observations. However, their visual prediction objectives do not explicitly distinguish task-relevant transitions from incidental scene changes. For whole-body policies, this distinction matters because task…
InJePA: Intention-First Latent Planning with Coupled JEPA World Models for Visual Navigation
Visual navigation requires goal directed planning informed by predictions of environment dynamics. Representative navigation world model methods propose complete action candidates before predicting consequences for selection or search updates. For each candidate, predictions…
EAGLE: Entropy-Aware Geometric Learning Enhancer for Partially Polygon-Supervised Object Localization
AI summaryEAGLE quantifies per‑class shape complexity with an entropy‑based metric and adds a lightweight post‑training head that refines edges and rescales localization scores, yielding about +0.9 mAP50‑95 on COCO for existing detectors with negligible inference cost.
Rethinking weakly supervised localization on chest radiographs: position nulls, protocol dependence, and PACE
Weakly supervised localization on chest radiographs is reported as a single overlap accuracy, and those numbers are not comparable across papers. We measure what one contains. Against position-null controls that keep a readout's own box and move only its position, class…
Your Self-Supervised Projection Head Captures Object Co-Occurrence Statistics
AI summarySelf‑supervised models encode object co‑occurrence statistics in their projection heads, while backbones learn categories. By shaping co‑occurrence patterns during pre‑training, category representations emerge faster, showing that leveraging co‑occurrence structure improves SSL learning.
Announced Breaks Separate Conformal Reliability from Frequency Calibration
AI summaryThe paper proposes a reset rule at announced boundaries that separates conformal reliability from frequency calibration, achieving minimax per‑time rate for independent scores. Experiments show it outperforms five adaptive methods across synthetic and real data.
Vocabulary-Incomplete Unsupervised Vision-Language Model Adaptation
AI summaryVI-UVA adapts vision‑language models to unlabeled data when only a subset of class names is known, using Label Space Reconstruction to build a latent label space and reduce bias toward known classes. LSR achieves state‑of‑the‑art accuracy on eight benchmarks.
Zero-Shot Tabular Learning via Hypergraph Reasoning with LLMs
AI summaryThis paper introduces TABHGR+, a zero-shot tabular learning framework that uses LLM-generated rules and hypergraph label propagation to model inter-row dependencies. It significantly reduces inference costs to a constant number of LLM calls while outperforming existing zero-shot and strong few-shot baselines in classification accuracy.
Improving Fake Profile Detection Through Multimodal Learning and Research-Guided Feature Integration
AI summaryMultimodal fusion of text, images, and metadata on TwiBot‑22 shows only marginal improvement over metadata‑only models; graph models perform worse under subsampled conditions. The study concludes that high‑quality metadata features are more valuable than complex multimodal or graph fusion.
miniTE: Diagnosing and Bridging the Execution Gap in Native NVFP4 Reinforcement Learning
AI summaryThis paper introduces miniTE, a framework that diagnoses and bridges cross-engine numerical discrepancies in native NVFP4 reinforcement learning for LLMs. It achieves stable training with near-BF16 performance while delivering up to 1.47x the throughput of standard NVFP4.
VERDICT: Visual Extraction-Result Decoder with Integrated Confidence and Traceable Evidence
AI summaryVERDICT attaches a Vision–Language Model to any extractor to localize evidence and calibrate confidence for both extracted values and evidence locations. Fine‑tuning the VLM boosts value AUROC to 0.914 and evidence AUROC to 0.906, enabling high‑precision coverage at 0.80 confidence.
Organizing LLM Inference for Aggregate Predictions from Behavioral Data
AI summaryThe paper investigates how to prompt LLMs for aggregate behavioral predictions without fine-tuning. It finds that evaluating summarized user histories in comparative prompts achieves the highest accuracy and lowest cost, nearly matching supervised baselines.
Factorization-Guided Training and Inference for Diffusion-Drafted Speculative Decoding
AI summaryFactorizes diffusion‑draft training into inclusion, weighting, and prediction losses, and throughput into mean accepted tokens per round and latency, to guide training and inference configuration. Achieves up to 4.14× speedup over autoregressive decoding on Qwen3 models, 7–9% better than prior baselines.
Beyond Prediction: A Component-Level Study of Concept Recovery in In-Context Learning
AI summaryComponent‑level analysis of in‑context learning shows that LLMs can predict accurately while failing to recover the whole concept; jointly specifying attributes and level constraints improves recovery, but it is fragile to unnecessary constraints, and joint supervision enhances both recovery and prediction.
SECURE-AI: Multi-Signal Uncertainty Quantification for Trustworthy Acoustic Cough Detection
AI summarySECURE-AI wraps a frozen MobileNetV3 cough detector with a multi‑signal uncertainty module (Monte Carlo Dropout, Temperature Scaling, Mahalanobis distance) and shows that Mahalanobis distance alone achieves AUROC 0.86 for separating true from false coughs, raising precision from 60 % to 80 % when rejecting two‑thirds of false positives.
Pack and Extract: An Effective Graph-Based Approach for Feature Selection
AI summaryThe paper introduces a graph-based feature selector that extends the Independent Dominating Set to a node packing model, prioritizing diverse features. On five genetic datasets, it improves predictive accuracy and feature stability compared to standard methods.
PeekKV: The First Content Token Already Knows Which Heads Matter
AI summaryPeekKV compresses LLM KV caches by peeking one content token after prefill to score attention heads and tokens, allocating cache budgets per head group without offline profiling. It achieves over 95% of full-cache accuracy at 1.5–10% memory budgets across benchmarks.
Fast Construction of Inner Sphere Trees Using Machine Learning
AI summaryML-IST learns to predict sets of inscribed spheres for inner sphere trees in a single forward pass, then refines them with differentiable coverage and projection. It halves or halves leaf-generation time compared to CPU, and achieves >55 FPS in physics engines, outperforming existing IST, GImpact, and convex decompositions.
A Group-Aware Comparative Study of Classical Machine Learning and Deep Learning Methods for Sensor-Based Human Activity Recognition
AI summaryThe paper compares four classical and four deep‑learning HAR models on UCI‑HAR and KU‑HAR using group‑aware cross‑validation. Deep learning only slightly outperforms classical models on UCI‑HAR; on KU‑HAR, LightGBM wins (0.857 vs. 0.828), and combined ACC+GYRO generally yields the best results.
ReTTT: Reflection-Driven Test-Time Training via Experience-Based Parametric Memory
AI summaryReTTT trains a frozen LLM agent during each episode by reflecting on its own interaction feedback and storing lessons in memory; the agent updates a LoRA adapter for policy and reflection. This method improves task success by up to 12.14 points over trajectory‑based test‑time training on ALFWorld, ScienceWorld, and SWE‑bench Lite.
How Data Scales in Agentic Reinforcement Learning: Laws and Synthesis Strategies
AI summaryThe paper decomposes agentic RL data into task, environment, and trajectory axes, finds scaling exponents ordering environment > task > trajectory, and uses these laws to optimize budget allocation, achieving up to 1.7× compute efficiency over naïve scaling.
Parameter misspecification in physics-informed neural networks: A closed-form audit and its limits
AI summaryPINNs trained with an incorrect parameter can still achieve low loss; this paper derives a closed‑form estimator for the parameter from the frozen network and shows its error correlates with the residual norm. In practice, the estimator detects most 1–2 % misspecifications but fails for very small errors or when the parameter also appears in the architecture.
DS-Net: A Dynamic Split Network for Edge-Cloud Collaborative Inference
AI summaryDS‑Net enables a single DNN to flexibly switch split points between edge and cloud, cutting memory use by 74.6% and switching latency to milliseconds while a hierarchical compensation mechanism improves accuracy and reduces latency by 42.5% on COCO at 10 Mbps.
Compaction-Aware Reinforcement Learning For Ascend Triton Kernel Generation
AI summaryATRL refines GRPO with a compaction-aware, segment-level objective, using cascaded compilation, correctness, and profiling feedback plus Ascend-specific knowledge to generate high-performance Ascend Triton kernels. It outperforms baselines on NPUKernelBench and KernelBench in accuracy and speed.
LatLM: Line-Parallel Language Models with Deferred Commitment
AI summaryLatLM allows line-parallel decoding with adaptive commitment, scoring eligible frontier tokens each round and deferring unready ones. It matches an autoregressive model’s accuracy while reducing decoding rounds by 25–41%, improving cross-line tasks from 78.5% to 96.5% accuracy.
DISCERN: Identifiable Disease Evidence for Generalizable Medical Multimodal Learning
AI summaryDISCERN learns a latent disease-state representation that is invariant to cohort selection, removing enrollment shortcuts. It improves state AUROC from 0.043 to 0.883 on synthetic shifts and increases cross‑database mortality AUROC by 0.036–0.035 versus empirical risk minimization.
Physics-Aware Self-Supervised Pretraining on Multi-Domain Raman Spectra
AI summaryPhysics-aware masked autoencoder pretrains a Raman encoder on heterogeneous unlabeled spectra, using intensity reconstruction, sparse Voigt peak decoding, and denoising. It improves label efficiency, robustness, and transferability on classification, component ID, concentration estimation, and denoising, especially with larger encoders and limited data.
Same Space or Not: Benchmarking Spatial Intelligence Through Cross-View Contradictions
AI summaryThis paper introduces Same Space or Not (SSoN), a benchmark for evaluating whether vision-language models can maintain coherent 3D scene representations by identifying cross-view contradictions. The results show that current models struggle significantly with this task, achieving low accuracy compared to humans, though performance improves with targeted supervision.
UltraData-Code: From Raw Repositories to Algorithmically Dense and Task-Oriented Code Data
AI summaryUltraData‑Code introduces a tiered data pipeline (L0‑L3) that filters, normalizes, selects algorithmically relevant code, and turns it into structured exercises, yielding ~400B L2 tokens and 150B L3 tokens. Using L2 improves EvalPlus pass@1 by 7.8pp, and adding L3 boosts it by another ~8pp.
A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences
AI summaryControlled LLM workflow combines reference retrieval, contrastive SHAP, reference-based reasoning and deterministic validation to explain risk score differences; it yields more consistent, evidence‑grounded diagnoses than an unconstrained LLM.
Geo-vision: Native-Resolution Geometry and Sampled Topology for Vehicle-Part Damage Classification
AI summaryGeo-vision classifies vehicle-part damage by fusing appearance, multiscale geometry, orientation, and sampled digital topology in a fixed representation, achieving 0.7577 AP—better than a YOLO11n baseline—while using fewer labels and modest CPU memory.
IMPULSE: Push First, Then Predict in V-JEPA 2 Feature Space
AI summaryIMPULSE introduces an active interaction strategy for V-JEPA 2 that uses a few short physical pushes to update latent world models. By incorporating this experiential memory and a Fourier pose head, the method reduces one-second feature-forecast error on novel objects by up to 16.6%.
Spatio-Temporal Detection of Photovoltaic Systems on Industrial Building Rooftops Using Satellite Imagery and Machine Learning
AI summaryThe paper builds a Random Forest model that detects industrial rooftop PV arrays using open-source Sentinel‑2 multispectral data combined with OpenStreetMap building outlines, and implements a web prototype that delivers up‑to‑date detections with quality flags for users.
Super-resolution as Spatial Downscaling: Coherence, Structure and Task Utility in Earth Observation
AI summaryLearned satellite super‑resolution can be viewed as spatial downscaling; across 735 tiles, lightweight models outperform bicubic by ~1.4 dB, recover ~30 % of high‑frequency spectral power, and change land‑use accuracy by up to ±3.1 pp, though they may amplify spurious changes.
FrameScout: Scouting Query-Relevant Frames for Long Video Understanding
AI summaryFrameScout selects query-relevant frames for long videos using a streaming selector that incorporates temporal context and a frame-query contrastive objective. It outperforms existing methods on three long-video benchmarks across six VideoLLMs, achieving state‑of‑the‑art results.
LLM Evolutionary Search Strategies Should be Compared by Tokens
AI summaryLLM evolutionary search is usually benchmarked by proposal counts, but this obscures real compute cost; using token counts shows that evolution’s apparent advantage disappears—evolution only outperforms independent sampling by ~5% in most cases, with a 23–28% margin on only one task.
SPELUNKA: A Framework for Finding Unanticipated Associations in Vision-Language Models
AI summarySPELUNKA automatically discovers unanticipated spurious associations in vision‑language models by generating captions, extracting entities, and using a frequency rule to detect over‑ or under‑representation in retrieval outputs. It achieves 0.93 AUROC across four datasets and outperforms baseline methods.
Weakly Supervised Discovery of Motion Styles for Compositional Robot Control
When a robot possesses more degrees of freedom than a task requires, multiple trajectories can achieve the same objective using distinct motion styles. Compositional robot control aims to preserve a selected style while simultaneously executing an independent task command,…
SPADE-KD: Peer-to-Peer Collaborative Learning under Heterogeneous Settings
AI summarySPADE‑KD is a fully decentralized distillation method for heterogeneous federated learning that adaptively weighs incoming model transfers by local loss benefit and selects peers by utility, achieving faster convergence and lower loss than naive peer‑to‑peer distillation.
TWIN-MoE: Two-stage Intra-Expert Imputation Network for Incomplete Multimodal Learning
AI summaryTWIN-MoE introduces a two-stage intra-expert imputation: pre-imputation creates context-conditioned estimates clipped by modality statistics, then post-imputation refines them using expert-conditioned aggregates. It achieves state‑of‑the‑art performance on four benchmarks with fewer parameters than existing MoE approaches.
Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis
AI summaryFusionRare fuses symbolic reasoning, deep learning, case retrieval, and LLM inference to prioritize rare disease diagnoses, achieving 51% Recall@1 for phenotype-only and 79% for phenotype‑genotype cases—outperforming DeepRare, GPT‑5.5, PhenoBrain, and LIRICAL.
Dynamical Kernel Attention: Estimating the Present from a Record of the Past
AI summaryDynamical Kernel Attention (DKA) derives self‑attention weights from a latent state that evolves under stochastic dynamics, bounding evidence from decaying past tokens and allowing the model to abstain when information is weak. This approach preserves standard attention’s parallelism and cost.
Do Knowledge and Reasoning Oriented Benchmarks Induce Stable Representation Signatures in LLMs?
AI summaryThe study probes five large LLMs for a stable internal signature tied to knowledge versus reasoning labels by permuting task labels and selecting the best layer per permutation. It finds no reusable signature: task labels are recoverable, but no consistent internal representation appears.
When Does Evidence-Graph Structure Predict Answer Correctness? Label-Free Abstention for Structured Retrieval
AI summaryThe paper proposes a training-free abstention method that uses the weighted cut between selected and unselected evidence in a graph to flag unreliable answers. On MemoryAgentBench it improves the precision‑coverage AUC from 0.736 to 0.785 and can identify a 71‑query subset with perfect precision.
Visual Causal-chain Bookmarking: Retrieval-Weighted Credit Assignment for Reinforcement Learning from Delayed Reward
AI summaryVCBM assigns more credit to causally relevant decisions in delayed-reward tasks, achieving 95% correct causal discovery in a watch-repair MDP in ~1,532 episodes—half the episodes of RUDDER—and modestly boosts LLM task success in AppWorld over LOOP.
Client-Aware Security Adaptive Early Exit Vision Transformers for Heterogeneous Federated Learning
AI summaryClient‑aware early‑exit ViTs in federated learning use client‑specific thresholds and adaptive routing based on entropy and confidence, achieving 88.7% CIFAR‑10 accuracy with 32% depth saved and 73.7% ImageNet‑1K top‑1 accuracy with 30% saving.
What Remains Without Test Labels? Re-evaluating a Graph-Level Anomaly Detection Benchmark
AI summaryThe study audits graph-level anomaly detection benchmarks, showing that using test labels for model selection boosts AUROC by ~5 points. Removing test-labels reduces performance gaps, revealing that many reported gains stem from label leakage rather than true algorithmic progress.
Alliance of Stochastic Routing and Reinforcement Fusion for Multi-Model Reasoning
AI summaryVL‑Scout uses Dirichlet‑based uncertainty priors, a KL‑regularized REINFORCE routing policy, and a GRPO/RLVR fusion policy to perform turn‑based, uncertainty‑aware multimodal model routing, achieving consistent gains on MMMU, MMMU‑Pro, OKVQA, and RealWorldQA while preventing premature policy collapse.
ToolRosetta: Surface-Form Invariance of Tool-Call Guards
Step-level guards for tool-using agents are evaluated on tidy JSON tool calls, but deployed coding agents act through execute_bash and execute_code tools whose payloads are strings. We introduce ToolRosetta, which re-renders every candidate action in the AgentDojo and AgentHarm…
OrTeN: Neural Inverse Graphics for Joint Terrain and Line-scan Camera Trajectory Reconstruction
AI summaryOrTeN jointly learns a textured height map and corrects orbital line‑scan camera trajectories using neural inverse rendering. On four planetary datasets it achieves terrain RMSEs of 2.35 m (Jezero), 2.82 m (Gale), 16.18 m (Olympus Mons) and 0.43 m (McGetchin).
Location, Location, Generation: Where You Ask Matters for Generative Engine Optimization
AI summaryGenerative engines tailor cited sources to the user’s declared location rather than network IP. ChatGPT localizes results to the city it is told, while Gemini uses the network country, altering recommended businesses and sources accordingly.
GLOW: General Linear Models Optimized with Ward's Method
AI summaryGLOW introduces a neuroimaging analysis method that builds a region hierarchy by minimizing General Linear Model error using Ward's clustering, combined with permutation testing for family-wise error control. It recovers spatially whole effects more reliably than voxel-based baselines, though at the cost of lower positive predictive value.
Interpretability of AI weather models via intermediate decoding
AI weather prediction models are often regarded as black boxes. Despite their impressive skill, there is little understanding of how the input data is processed during inference to produce the forecast. This limitation arises from the use of high-dimensional latent…
Looped GNNs Beyond Fixed Depth: Graph Size Generalization and Parameter-Efficient Adaptation
AI summaryTrains a GNN on small graphs, then loops the same learned update beyond training depth to handle larger graphs. Shows when looping works, and introduces Loop‑Freeze‑Steer (LFS) that freezes the processor and learns a tiny controller, greatly reducing parameters while improving accuracy on big graphs.
Graph-Regularized Synthetic Control for Stable Counterfactual Estimation under Donor Redundancy
AI summaryGraph-Regularized Synthetic Control adds a graph Laplacian penalty to classical synthetic control, encouraging coherent weights across similar donors. It yields more stable, interpretable counterfactuals while matching pre‑ and post‑treatment accuracy, especially in donor‑redundant settings.
Do LMs Know What Variables Do? Interpreting Code Representations via Variable Roles
AI summaryThis paper investigates whether language models and sparse autoencoders capture programming variable roles, introducing a multilingual corpus with static annotations. The results show that models recover these roles successfully, but causal interventions often degrade code correctness.
Prefix-Tuning for Arbitrary Output Sequences on Pretrained Transformers
AI summaryThe paper shows that for any pretrained autoregressive Transformer and any target output within the model’s range, a prefix can be constructed by solving linear systems to force the model to generate that output; it also proves this condition is necessary.
On the Selectivity of Generative Models in Structure-Based Drug Design
AI summaryThe paper introduces SELECTBENCH, a benchmark for assessing selectivity in generative structure‑based drug design. It evaluates nine models on three real‑world tasks, finding that current methods, even those trained for selectivity, only modestly discriminate off‑targets.
Data-Driven Resource Aggregation for Scalable Digital Twin and Simulation Model Generation
AI summaryThe authors propose a data-driven method that uses inverse reinforcement learning on event logs to automatically aggregate similar manufacturing resources. The approach significantly reduces simulation model complexity while preserving performance accuracy.
Recording-Aware Evaluation of Multimodal OT Intrusion Detection
Machine-learning-based intrusion detection systems (IDSs) for operational technology (OT) are commonly evaluated on benchmark datasets in which normal and attack data are collected in separate recordings. In such settings, high performance may reflect recording-associated…
Reducing double-dipping in single-cell topic models with three-way count splitting
Topic models learn gene programs from single-cell RNA sequencing (scRNA-seq) data, and the same counts are then used to test which genes and donors are associated with those programs. Count splitting is the standard remedy for this reuse, but it has not been tested when an…
The Weight of Teamwork: Does Communication Structure Help LLM Scientists Search for Scientific Laws?
Team performance is often reported as one delivered score, even though that score is produced in two stages: a system generates candidate answers and a selection rule returns one of them. We use symbolic regression as an observable scientific-search setting and compare…
Inside the Loop: A Mechanistic Study of Weight-Tied Transformers on Depth-Bound Algorithmic Tasks
AI summaryThis paper investigates how weight-tied looped transformers solve multi-step algorithmic tasks by reusing layers across iterations. It finds that looped models act as iterative program learners that outperform identically-trained untied transformers of the same effective depth.
Continual Test-Time Learning with Evolution Strategies
AI summaryThis paper introduces Evolution Strategy Reinforcement Learning, a framework that uses a population of perturbed policies to generate stable, label-free supervision for continual test-time adaptation in language models. The method improves average accuracy by up to 31.7% and significantly reduces catastrophic forgetting during continual learning across diverse domains.
Sequence-to-Sequence Neural Surrogate Modeling of Transient Responses in High-Speed Links with Variable Design Parameters
This paper presents sequence-to-sequence (seq2seq) neural surrogates for high-speed transient response modeling and signal integrity analysis in fixed and variable design parameter spaces. To identify the structural requirements for prediction accuracy, multiple neural…
lepidoptera: A Domain-Agnostic Engine for Causal Simulation with Mixed Variable Types, Vine Copula Dependence, and Individual-Level Counterfactual Propagation
lepidoptera is an open-source R and Python package for simulation from structural causal models with heterogeneous variable types. The package is built around a typed mechanism architecture, which separates the causal graph from the generating rule at each node and exposes each…
Budgeted Bayesian Bandits for Future Utilization
AI summaryBudgeted Bayesian bandits study exploration with a hard budget before exploitation. The paper tightens prior 1/3 and 4/27 bounds, presents a 0.43-approximation algorithm, and an efficient 0.63-approximation algorithm for bounded exchangeable rewards, matching LP relaxations.
Margin Certificates for Voting Classifiers under Mixing Dependence
AI summaryThis paper derives uniform margin certificates and generalization bounds for voting classifiers under stationary mixing dependence. The results provide non-vacuous guarantees and valid confidence intervals on dependent time-series data where standard i.i.d. methods fail.
Bootstrap Stochastic Interpolant Tree for Inference-time Search Under Sparse Reward
AI summaryBootstrap Stochastic Interpolant Tree (BSIT) decouples child‑node sampling from parent transitions, enabling diverse trajectory rollouts and wider exploration in inference‑time search. Experiments show BSIT converges faster to target distributions and outperforms DDPM‑based tree samplers in rare‑reward tasks.
On Reparameterizing the Score Function
AI summaryThis paper introduces History-Reparameterized policy gradients to alleviate the credit-assignment bottleneck in autoregressive sequence models by backpropagating through the conditioning history. The method uses an optimal control-variate estimator to reduce gradient variance and improves performance on sparse-reward tasks.
Candidate-dependent residual routing trades top-of-list accuracy for recall in sequential recommendation
AI summaryCandidate‑dependent residual routing adds a lightweight module to a self‑attentive sequential recommender, keeping NDCG@10 unchanged while boosting recall by 1.8% at the expense of a 1.1% drop in MRR, mainly moving hits from positions 1–5 to 6–10.
Certified neural approximations for parabolic obstacle problems
Residual-trained neural surrogates carry no error guarantee, a limitation for free-boundary problems such as phase change and contact. We derive guaranteed error bounds for neural approximations of parabolic obstacle problems in first-order state--flux--multiplier form. The main…
Hidden Trajectories Don't Lie: Correcting Corruption Drift for Test-Time Adaptation
AI summaryDrift‑TTA aligns hidden layer trajectories of a frozen backbone via low‑rank adapters, learning from clean or corrupted inputs. On CIFAR‑100‑C severity 5, it boosts ViT-B/16 accuracy by 13.3 pp and reduces ECE to 9.53 %.
Training-Free Prompt Selection for Efficient Reinforcement Learning With Partial Rollouts
AI summaryThis paper introduces LIFT, a training-free prompt selector for reinforcement learning that evaluates data using only partial reasoning rollouts. LIFT reduces selection token costs by up to 9.1 times while maintaining competitive downstream reasoning performance.
Censored Flow Matching for Synthetic Time-to-Event Generation
AI summaryCensored Flow Matching (CFM) is a flow-based generative method for time‑to‑event data that handles censoring via pullback sampling, yielding an EM algorithm that converges to the true event‑time distribution and enables realistic synthetic data and accurate density estimation.
Preserving Reasoning Diversity in Reinforcement Learning with Verifiable Rewards via Adaptive Replay
AI summaryThe paper analyzes how limited on‑policy rollouts in RLVR suppress semantically distinct reasoning modes and introduces TAR, a Targeted Adaptive Replay that stores diverse base-model solutions and replays them only when their usage drops, improving both performance and diversity.
Encoder-Free Multimodal LLMs Learn Implicit Encoders
AI summaryEncoder‑free multimodal LLMs (Gemma‑4 12B, Inkling‑Small) learn “implicit encoders” in their early layers that transform audio and visual inputs into representations largely independent of cross‑modal context. Layer probing shows these implicit encoders match the behavior of standard dedicated encoders, and early‑modality encoding alone matches joint‑encoding accuracy.
Extending Myerson's Optimal Auctions to Correlated Bidders via Neural Network Interpolation
We aim to design revenue-maximizing single-item auctions that are deterministic, strategy-proof, and ex post individually rational --- the quintessential fundamental model for optimal mechanism design. While Myerson's seminal work solved this model for independent bidders,…
CoCoNet: Convex Conformal Network for Multi-Dimensional Uncertainty Regions
AI summaryCoCoNet learns convex conformity scores for multivariate regression by optimizing prediction‑region efficiency on held‑out data, then applies split‑conformal calibration to keep coverage guarantees. Across 30 tasks, the ICNN variant achieves the highest efficiency, ranking first on 22 of 30 tasks.
Can This LLM Annotate My Data? Estimating Zero-Shot LLM Performance
AI summaryThe paper constructs 461 synthetic text classification datasets and shows that zero‑shot LLM accuracy can be predicted from dataset features; internal log‑probabilities outperform confidence verbalizations for predicting performance.
Cortex: A Fixed-Point Theory of Governed Coding Agents - Technical Report
AI summaryCortex adds a supervisory layer that plans, edits, and validates code from a large language model. The paper proves that, under sound, monotone validation rules, exhaustive or fair scheduling will converge to the least fixed point of the requirement lattice, and provides a stochastic bound on expected iteration time.
Dynamic Neural Optimal Transport for Subset Alignment
AI summaryDynamic neural optimal transport learns a terminal distribution within a target’s support while enforcing a density‑ratio bound [0, c] to select relevant subsets. Applied to PU learning, unpaired translation, and trajectory inference, it outperforms fixed‑terminal approaches.
Dataset Inference for Data Provenance and Privacy Auditing in Tabular Foundation Models
AI summaryThe paper proposes the first dataset‑inference attack for tabular foundation models, combining multiple black‑box signals to detect if a suspect dataset was used for pre‑training. It achieves up to 0.99 ROCAUC on state‑of‑the‑art TFMs, enabling privacy and provenance auditing.
BanglaShield: A Script-Robust Safety Classifier and Benchmark for Bangla Native, Romanized, and Code-Mixed Prompts
AI summaryBanglaShield provides a graded safety classifier for Bangla prompts in native, romanized, and code‑mixed forms, achieving 87.8 % macro‑F1 on native script and 84.4 % on romanized, outpacing large LLMs by over 40 points while using ~130× fewer parameters.
An important Bottleneck in LLM Optimizers: Principle-Level Local Optima
AI summaryThe paper defines “principle-level collapse” where LLM optimizers get stuck exploring only a few design principles despite generating varied candidates, and shows that common diversity tricks rarely help. It proposes a diagnostic benchmark and highlights the need for online collapse detection and better principle generation.
Inducing Fourier Geometry: A Smoothness Prior Imposes Approximate Translation Equivariance in Multimodal Stellar Embeddings
AI summaryApplies a graph‑Laplacian smoothness prior to multimodal stellar embeddings, inducing a Fourier-like, Toeplitz structure along physical parameters; this improves linear probe accuracy and label efficiency without requiring explicit symmetry assumptions.
Constructive Universal Approximation by Infinite-Dimensional Neural Networks via Gaussian-Weighted Ridgelet Transform
AI summaryThe paper extends the ridgelet transform to infinite-dimensional Hilbert spaces using Gaussian-weighted directions, yielding explicit coefficient measures and reconstruction formulas. It proves uniform approximation of function‑valued neural networks for non‑polynomial activations, with a dimension‑free Monte‑Carlo error rate \(O(N^{-1/2})\).
Predicting CVSS and CWE from CVE Descriptions with Fine-Tuned Small Language Models
AI summaryFine‑tuned small language models can predict CVSS vectors, severity tiers, and CWEs from CVE descriptions. A 3.8B Phi‑3.5‑mini model trained in 2.5 h on an Apple M5 Pro attains ~70 % accuracy on severity and CWE, matching official vectors in 48.8 % of 2024 CVEs and surpassing TF‑IDF baselines.
Torch-DPhys: A Torch-Native, MOOSE-Compatible Framework for Differentiable Physics Simulation and Hybrid Machine Learning Integration
AI summaryTorch‑DPhys is a PyTorch‑native framework that implements differentiable thermal‑hydraulic and continuum‑mechanics models from MOOSE, enabling CPU/GPU execution, sparse solves, and analytic or adjoint sensitivities. In a molten‑salt reactor test it matches SAM‑Lite with <1.13×10⁻⁵ error and recovers pump head and heating power with <0.06 % error in a parameter‑identification run.
Muon Operates Beyond the Edge of Stability
AI summaryMuons optimizer satisfies the Flat Minima Hypothesis: sharpness correlates with generalization (r = 0.45–0.69). It stays stable above the classical edge‑of‑stability threshold, indicating a different stability regime than Adam or RMSprop.
Learning Administrative Boundary-Aware Location Encoders via Reverse Geocoding
AI summaryThe paper trains location encoders with reverse‑geocoded text to explicitly encode administrative boundaries, improving downstream geospatial tasks. A new RevGeoRQ model achieves state‑of‑the‑art performance and offers more interpretable, residual‑quantized geographic representations.
RATION: Towards Quantitative Risk-Based Access Control for AI Agents
AI agents increasingly interact with external environments to complete complex user tasks. These interactions can lead to unexpected behavior and expose agents to security threats such as indirect prompt injection. Current agent security systems largely protect agents by…
Latent State Utilization in Probabilistic Circuits
AI summaryLatent states in probabilistic circuits increase capacity but add little likelihood. In HCLTs on ImageNet32, mutual information shows slow growth of learned dependence, and early posterior concentration guides better initialization and tree selection, yielding measurable performance gains.
Morpheus-2.0: Multiscale Probabilistic Forecasting with Dual-Axis Attention
AI summaryMorpheus‑2.0 trains a single encoder‑only model that jointly attends across variables and time on a variable‑time grid, sampling horizon lengths during training. The multiscale version handles multiple resolutions, achieving 0.646 relative MASE and 0.455 relative CRPS, the best reported.
PAC Learning with Query Feedback
AI summaryThis paper introduces the query dimension, a new combinatorial parameter that characterizes the sample complexity of PAC learning with query feedback. The authors prove matching upper and lower bounds up to logarithmic factors for this learning setting.
Power-Law Correlations Without Apparent Temperature-Tuned Criticality in Fixed-Tree Hierarchical Languages
AI summaryThe study shows that on a finite-depth context-free grammar tree, temperature scaling of softmax during autoregressive sampling only changes the magnitude and exponent of a persistent power-law mutual-information decay; no phase transition or long-distance plateau appears within the tested sizes.
Uniform convergence at optimal rate and scale
AI summaryUniform convergence for real-valued function classes now attains the optimal sample complexity of \(O(d/\varepsilon^2)\) at the optimal scale \(γ\), matching the binary VC case. This is achieved via margin‑ε‑nets, extended Haussler covers, and an aligned chaining argument.
What Matters for NVFP4 Training? A Comparative Study of Low-Precision Pretraining Recipes
AI summaryNVFP4 training of 8B and 30B MoE language models to 1T tokens is studied, showing a recipe that combines unbiased gradient estimation, selective high‑precision layers, and partial tensor de‑biasing. The final method achieves the smallest loss gap to BF16 at this scale.
The Certification Exponent: Instance-Optimal Sequential Certification of Policy Improvement from History Dependent Logs
AI summaryThe paper shows that the optimal number of episodes needed to certify a policy’s improvement from history‑dependent logs is \(\Theta(\log(1/\delta)/K^\star)\), where \(K^\star\) is the minimum KL divergence to a “bad” law. A plug‑in Kelly certifier attains this bound, and a backward‑induction “coverage” test identifies whether certification is impossible, finite‑rate, or constant‑time.
Measuring the Task Value of Precision Upgrades in Closed-Loop Agents
AI summaryA low‑precision agent can call a full‑precision fallback at selected decisions, guided by a precision router that uses model disagreement. The paper shows that, across Qwen3 agents, the signed task benefit of a BF16 fallback is mixed: action‑only rescues and damages largely cancel, while replacing both action and continuation yields a positive net gain in some lower‑bit settings.
Exact Risk Certificates for Pairwise-Judged Reranking
AI summaryThe paper derives exact worst-case nDCG loss bounds when a ranking is replaced by a fixed proposal using pairwise relevance judgments, showing binary gains attain this loss and enabling efficient verification via up to K+1 closure problems. Experiments show the approach reduces comparison effort by ~2.3 per query while maintaining guarantees.
Blackbox Performance Analysis of LLM-based Network Control
AI summaryLENS audits black‑box LLM‑enabled network controllers by extending deep reinforcement learning to search for performance gaps and hard inputs. It discovers significant shortcomings in LLM‑based traffic engineering and GPU allocation, matching the effectiveness of white‑box analyzers for hand‑crafted heuristics.
Can Agents Keep Secrets They Were Never Told? Benchmarking Latent Privacy Risks Across Sessions with MemJoin
AI summaryAgents use long‑term memory to combine user info across sessions, risking unintentional disclosure of sensitive relations. MemJoin benchmarks 288 cases, showing that even when individual sessions hide the relation, memory systems can reveal it, and attribute visibility checks are insufficient.
Revisiting Smoothed Online Learning: Competitive Ratio, Dynamic Regret with Switching Cost, and Best of Both Worlds
AI summaryThe paper presents improved competitive ratios for smoothed online learning and optimal dynamic regret with switching cost, achieving \(1+2/\sqrt\lambda\) for quadratic costs and \(O(\sqrt{T(1+P_T)})\) regret; it also introduces UDNP for best‑of‑both‑worlds guarantees without bounded gradients.
Learning from the Right Regions: Class-Conditioned Explanation Guidance for Medical Imaging
AI summaryClass-conditioned explanation guidance trains multi‑label medical image classifiers to focus on the correct regions per abnormality, penalizing evidence outside class‑specific annotations. Across EfficientNet, ViT, and Swin models, it boosts All‑Saliency Precision from 0.14–0.10 to 0.40–0.57 with minimal accuracy loss.
A decomposition-integration framework coupling singular spectrum analysis with a volatility-adaptive LSTM for stock price forecasting
Financial time series exhibit multi-scale, high-noise, and non-stationary volatility dynamics. Decomposition--integration frameworks built on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) have advanced this field, yet existing CEEMDAN hybrids…
Learning volatility-adaptive forgetting via GARCH-modulated LSTM for stock price forecasting
Stock price forecasting is hindered by the heteroscedastic nature of financial time series, in which turbulent and tranquil regimes alternate. Although the Long Short-Term Memory (LSTM) network is widely used for this task, its gates are oblivious to such volatility variation,…
MSCR: Jointly Balancing Utilization Across Modalities and Discovering Synergistic Information
AI summaryMSCR introduces a regularizer that simultaneously balances modality use and boosts synergistic fusion in multimodal learning. It outperforms MCR on five datasets, improving conditional recovery and cross‑modal attribution while remaining effective with Transformer fusions.
Prototype-Guided Optimization of Quantum Data Embeddings with Linear Cost
In supervised quantum machine learning with classical data, classification performance critically depends on the geometry induced by quantum data embedding. Existing data-adaptive approaches rely on pairwise fidelity evaluations between samples, which scale as $O(B^2)$ for $B$…
Cross-Dimension and Cross-Modality Relational Tensor Representation Learning for Multimodal Health Data Modeling
AI summaryRTR jointly learns dimension‑specific tensor representations and cross‑dimensional/modality relationships using relational bridge matrices, then fuses factor‑level and relationship‑level features. It outperforms tensor, fusion, and Transformer baselines on neurophysiological datasets.
Margin-Based Generalization for Quantum Machine Learning with Classical Data Encoding
Understanding generalization is essential for assessing the practical viability of quantum machine learning. In many applications, quantum models operate on classical data that must first be encoded into quantum states, making the encoding part of the effective hypothesis class.…
Transport What Shares the Conduction Pathway: Normal-Anchored SPD Geometric Transport for Source-Free Patient Adaptation of Holter and Patch ECG
AI summaryNAGT adapts ECG classifiers by estimating a symmetric positive-definite map from the first 5 min normal beats and applying it only to normal and supraventricular prototypes. It raises macro‑F₁ from 34.5 % to 48.7 % on 493 patients, with 5.8 ms per record.
Transformers as Adaptive Kernel Learners across Tasks for In-Context Learning
AI summaryTransformers learn adaptive anisotropic kernels in in‑context learning. The paper shows that, for low‑rank linear and multi‑index nonlinear regressions, the learned attention matches oracle kernels, and provides a theory linking transformer predictions to oracle kernel estimators.
Near-Optimal Regret for Semi-Bandit Learning of Monotone Stochastic Problems
AI summaryThe paper presents a semi‑bandit learning algorithm for a broad class of monotone stochastic problems, achieving $\tilde{O}(\sqrt{nkT})$ regret for weakly monotone and $\tilde{O}(\sqrt{nT})$ for strongly monotone cases, matching the optimal $\Omega(\sqrt{nT})$ lower bound.
Generative Retrieval for Reverse Target Fishing with a Proteome-Scale Benchmark
AI summaryGenerative Retrieval for Reverse Target Fishing (GenFish) encodes protein language model features into shared semantic identifiers and autoregressively predicts them from a molecule’s fingerprint, avoiding false negatives from unrecorded pairs. On ProTargetBench, GenFish outperforms discriminative baselines in full‑catalogue retrieval of 20,431 proteins.
Measuring Empowerment from Offline Data for Intrinsically Motivated Exploration
AI summaryThe paper introduces a method to estimate empowerment solely from unlabeled offline data, training a policy and occupancy measure jointly. This estimator yields intuitive empowerment scores and boosts exploration in online goal-conditioned reinforcement learning.
Private Evaluation of High-Dimensional Functions
AI summaryThis paper introduces generic differentially private black-box evaluators for high-dimensional functions via a reduction to private quasi-concave optimization. The method guarantees outputs within the convex hull of local function evaluations while improving data overhead for user-level optimization.
Write Once, Check Many: Reusable Ledgers for Post-Hoc Policy Compliance Auditing of Agent Trajectories
AI summaryThe paper introduces a write‑once, check‑many ledger system that records agent trajectories in a policy‑independent format, enabling efficient post‑hoc compliance checks. Experiments show ledger readers outperform trajectory readers, achieving higher AUC with far fewer tokens.
VBO-MI: A Differentiable Black-Box Optimization Framework via Variational Mutual Information Estimation
AI summaryThis paper introduces VBO-MI, a fully gradient-based black-box optimization framework that uses an actor-critic architecture and variational mutual information estimation. Experiments show it achieves competitive or superior performance compared to traditional Bayesian optimization baselines across high-dimensional and complex tasks.
CoTs remain monitorable when being unfaithful requires computation: Replicating and Extending the Necessity Effect Across Eleven Models
AI summaryThis paper replicates and extends the necessity effect across eleven language models, demonstrating that chains of thought remain faithful when tasks require complex computation. Results confirm that models struggle to conceal complex cue-following in their reasoning traces, supporting chain-of-thought monitoring for safety.
TideKV: Delayed KV Eviction via Layer-Wise Chunked Prefill for Memory-Efficient Long-Context LLM Inference
AI summaryTideKV introduces a layer-major delayed eviction schedule for long-context LLM inference that processes prompt chunks within each Transformer layer before evicting cache. This reduces peak prefill memory by up to 83.89% on A100 at 64K context while preserving full-prompt-aware selection quality and final cache size.
HiPPO Functional Memory: Coordinate-Queryable Attention for Irregular Time Series
AI summaryThis paper introduces HiPPO Functional Memory, an attention operator that compresses irregular time series into a reusable, coordinate-addressable functional state. The method improves reconstruction, extrapolation, and prediction across diverse irregular time-series benchmarks while enabling efficient query evaluation without reprocessing history.
BEM: Learning Bregman Divergences for Concept Erasure in Flow Matching Models
AI summaryThe paper introduces Bregman erasure matching (BEM), which learns an optimal Bregman divergence via bi-level optimization to remove harmful concepts from flow matching models. BEM outperforms existing baselines in erasing targeted concepts while better preserving unrelated generations.
Denoising Implicit Variational Inference
AI summaryThis paper introduces Denoising Implicit Variational Inference, which applies denoising score matching to directly estimate marginal scores in semi-implicit variational families. This approach outperforms existing implicit variational inference baselines while reducing gradient estimation costs.
What "Form'' of Uncertainty Does Conformal Prediction Quantify?
Conformal prediction (CP) is a post-hoc calibration method that returns prediction sets with distribution-free coverage guarantees, and the size of the output set is often interpreted as a notion of uncertainty. We show that the difference in set size between full and split CP…
TopoCompile: Molecular Generation with Exact Global Count Constraints
AI summaryTopoCompile combines an autoregressive Transformer with a dynamic-programming compiler to generate molecules that satisfy exact global count constraints, such as elemental composition and ring content. It achieves 100 percent constraint satisfaction and superior validity across four benchmarks without requiring model retraining.
ProbePO: Sparse Probes, Redistributed Credit for Critic-Free LLM Reinforcement Learning
Reinforcement learning from verifiable rewards assigns one terminal reward to an entire reasoning trajectory, obscuring how individual reasoning steps contributed to the outcome. GRPO broadcasts a group-normalized advantage to every response token, while finer-grained approaches…
Reconstruction of Unobserved Core Electron Density in Tokamak Plasmas with a Neural Operator Transport Surrogate and Uncertainty Quantification
AI summaryUsing a Fourier Neural Operator surrogate of a 1‑D transport solver, the paper reconstructs tokamak core electron density profiles from limited reflectometry/interferometry data, providing realtime‑compatible gradient‐based inference and uncertainty estimates that outperform conventional analytical fits.
Rethinking EEG Spatial Super-Resolution as Dynamic State Decomposition and Filtering
EEG spatial super-resolution aims to recover high-density neural signals from sparse electrode observations. Existing methods have achieved substantial progress through spatial modeling, generative reconstruction, and temporal representation learning. However, the current EEG…
Contrastive Hypergraph Source-free Domain Adaptive Object Detection in Adverse Weather Conditions--Extended Version
Adverse weather conditions such as fog, rain, snow, and low illumination introduce structured visibility degradations that severely impair object detectors trained on clear-weather data. When source data are inaccessible due to privacy or trans- mission constraints, source-free…
Structured Latent Dynamics for Long-Horizon Embodied World Models
Reliable long-horizon prediction remains a central challenge for embodied world models. Although recent generative models can synthesize visually plausible action-conditioned futures, their latent dynamics are typically learned through unconstrained transition modules, which may…
From Score to Diagnosis: Search-Independent Post-Diagnosis for Symbolic Regression
Symbolic regression (SR) solvers employ diverse search mechanisms, yet most rely on the same low-information scalar fitness. Such scores provide little guidance on why a skeleton fails or how it should be revised. Recent methods develop new methods to address this bottleneck.…
A Multi-Step Inertial ADMM Algorithm for Federated Learning
AI summaryMiFedADMM uses multiple past server iterates to form a higher‑order extrapolation in ADMM for federated learning, accelerating convergence. It provably converges (O(1/K) rate) and outperforms existing methods on synthetic and real datasets.
Write Once, Query Many: Evolving Compact Reusable Memory for Long Videos
AI summaryVideo‑EMem encodes long videos once into a compact memory of text and sparse images, then retrieves evidence for new queries. It improves Qwen3‑VL‑30B accuracy by ~2.6% on three QA benchmarks while cutting storage 99.86% and per‑query latency.
Accelerating Finite-Sum Bilevel Optimization with Fully First-Order Oracles
AI summarySICBO is a fully first‑order algorithm for nonconvex‑strongly‑convex bilevel problems with finite sums, achieving a gradient‑norm error ε with \(\tilde{O}(N+\sqrt{N}\kappa^{7/2}\varepsilon^{-2})\) component‑gradient calls, optimal up to log factors for fixed κ.
Theory-Guided Instance-Dependent Early Stopping via Second-Order Loss Dynamics
AI summaryThis paper introduces Theory-Guided Instance-Dependent Early Stopping (TG-IES), which establishes a theoretical foundation for second-order loss dynamics to identify and skip stable training examples. TG-IES saves over 50 percent of backpropagation instances on CIFAR benchmarks with minimal accuracy loss.
From Stories to Features: Scaling Concept Vectors for AI Alignment & Interpretability
AI summaryThe authors propose generating feature vectors for language model safety and alignment by taking the mean activation difference between synthetic concept-antagonist story pairs. Scaling this approach to 1,036 concepts across multiple models, they show these vectors generalize to detect concepts, reveal shared geometric properties, and causally steer model behavior.
No-Regret Revenue Maximization in Sequential Market Making Under Inventory Constraints
Market makers keep trading efficient by setting bid and ask prices for buyers and sellers. In sequential market making, trades happen independently based on price thresholds, requiring the market maker to maintain balanced, low-risk inventory over T rounds. This paper studies…
Representations Have Trajectories: Characterizing Temporal Representation Dynamics in Neural Networks
Recent work has shown that deep neural networks exhibit an implicit low-rank simplicity bias. In this paper, we investigate the questions that naturally arise from this observation. What are the temporal consequences of low-rank simplicity bias during optimization? How does…
ASR Personalization in Pathological Voice: Accuracy, Efficiency, and Duration Effects
Automatic Speech recognition (ASR) struggles with Pathological speech due to persistent speaker-specific mismatch. In clinical and accessibility scenarios, only limited target-speaker data is typically available, making full model retraining impractical. We present a…
DeliMF: Gated Cross-Attention Fusion of Structured Time Series and Clinical Text for Early Intensive Care Delirium Prediction
Delirium affects a large fraction of intensive care unit (ICU) patients and foreshadows longer stays, lasting cognitive decline, and higher mortality. Flagging the patients who will become delirious within the first day of admission would give clinicians a window to intervene.…
Conditional Weight of Evidence for Gradient-Boosted Credit Models: Estimation, Additivity and Its Relation to Adverse Action Reasons
AI summaryThis paper introduces conditional weight of evidence for gradient-boosted credit models to assess a characteristic's impact given an applicant's full record. The proposed tree-based estimators accurately recover applicant-specific credit information and complex feature interactions missed by traditional marginal measures.
Training with Ciphers to Enhance In-Context Terminology Learning For Translation from Rare Languages
AI summaryThis paper introduces RareCipherFT, a fine-tuning method that replaces rare translation terms with ciphers during training to force LLMs to copy from context. The approach improves in-context terminology translation by 5 to 15 points across multiple rare language pairs without degrading overall translation quality.
AdamPM: Memory-Efficient LLM Optimization via Moment Pooling
AI summaryAdamPM reduces Adam’s memory footprint by pooling second-moment entries into groups, keeping a single shared mean per group. With group size s, memory drops to 1/s of Adam’s, while preserving accuracy and speed on LLM pre‑training and fine‑tuning.
Severity-Channel Conditioning: An Injection-Level Study for Consumer-Device Deblurring
Motion-deblurring models lose substantial accuracy when transferred from professional cameras to smartphone-derived imagery. This paper studies whether a severity descriptor, partly derived from the paired sharp target, helps adaptation on the iPhoneBlur consumer-device…
TCMScience: An Autonomous Scientist for Traditional Chinese Medicine
AI summaryTCMScience is an autonomous research agent for Traditional Chinese Medicine that uses a trusted execution kernel and evidence contracts to coordinate heterogeneous data sources. The architecture prevents unsupported scientific claims, reducing erroneous outputs to zero in evaluations while correctly retaining valid conclusions.
How Much Do Activation Monitors Benefit from Original Model Computations?
AI summaryThis paper investigates whether internal activation monitors rely on original model computations to detect undesirable behavior. The authors find that while most monitor performance can be recovered without native traces, access to original computations provides a small but consistent predictive advantage.
Uncovering Semantic Hierarchies in Text-Attributed Graphs via Variational EM-based LLM-GNN Synergy
AI summaryThis paper introduces SHiFT, a framework combining Large Language Models and Graph Neural Networks via Expectation-Maximization to learn coarse-to-fine semantic hierarchies in text-attributed graphs. Experiments show that SHiFT achieves state-of-the-art performance while generating high-quality, human-readable taxonomies.
Modeling Uncertainty Propagation in 3D Morphable Models
AI summaryCLUE maps 3DMM parameter variance to per‑vertex uncertainty with a closed‑form formula, reducing computation. It improves mesh and uncertainty accuracy in training and inference, achieving the best uncertainty‑error correlation among baselines.
EdgeKernelBench: Unified LLM-Driven Kernel Generation for Edge Backends
AI summaryEdgeKernelBench provides a unified IR and hierarchical benchmark for generating LLM‑driven kernels on heterogeneous edge devices, revealing a 64.5% correctness but only 25.5% performance. Layered Edge Kernel Agent (LEKA) improves fast execution to 43.5%, outperforming direct and Claude Code.
Good Judgment, Bad Advice: Decision-Model Controllers Raise Agent Cost Without Improving Success
Agent scaffolds increasingly delegate small decisions (whether to stop, verify, or change approach) to cheaper models. We ask whether a typed decision model, Jev 1.13, consulted only at deterministic decision points, makes a Claude Sonnet 5.5 terminal agent cheaper without…
Lightweight Machine Learning for Classifying Disaster-Related Social Media Posts
Classifying disaster-related social-media posts into humanitarian information labels is challenging because individual messages are typically short and context-limited, humanitarian categories can overlap semantically, and the CrisisBench label distribution is strongly…
Can AI agents make law? Evaluating AI legal relevance and creativity
AI summaryThis paper evaluates LLMs drafting a novel constitution for AI communities via multi-character debates. The agents reliably form foundational legal rules and generate AI-specific proposals, though their creativity remains limited to recombining existing human legal concepts.
What RL Brings to Reasoning: A Few Decisions the Base Model Would Not Make
AI summaryThe authors investigate what reinforcement learning adds to reasoning base models by swapping tokens between them during generation. They find that borrowing the RL model's choices at just four percent of token positions recovers over ninety percent of its reasoning performance gains.
Knowledge Diversion: A Novel Paradigm for Pre-training Decomposable Image Generation Models
Large-scale pre-trained models have become foundational to modern machine learning, yet conventional pre-training objectives primarily optimize source-domain performance without explicitly considering downstream requirements. Consequently, transferable and task-specific…
MM-LeJEPA: Correlated Gaussian Regularization for Multimodal Joint-Embedding Predictive Architectures
LeJEPA derives an isotropic Gaussian embedding target by minimizing, under its task assumptions, a Fisher information criterion that controls the task-averaged squared prediction bias. This enables unimodal self-supervised learning without reconstruction decoders,…
Contrastive Adversarial Training for Robust Graph Neural Networks under Label Poisoning
Graph Neural Networks (GNNs) are effective for modeling relational data but are vulnerable to label poisoning, where a small number of corrupted training labels can propagate errors across the graph via message passing. Despite this risk, defenses against label poisoning remain…
Geometry-Aware Interpolation for Inverse Problems
Recovering missing visual information from incomplete or degraded images remains a challenging problem, particularly when spatial variations and complex image structures affect reconstruction accuracy. Although Spatial Transformer Networks (STNs) facilitate spatial feature…
Native Probability Outputs for Clinical Triage: An Evaluation of Calibration and Decision Accuracy
Patient-facing triage requires an appropriate care recommendation and recognition of when information is insufficient. We evaluate whether Jev, a decision model that returns probabilities over predefined actions, supports both judgments. Using CARE-Bench, we compare Jev with…
Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling
Meteorological downscaling is crucial for high-resolution regional climate forecasting and disaster early warning. While neural operators have emerged as a promising paradigm for modeling complex spatiotemporal mappings, existing frameworks often struggle with spherical manifold…
When Do Repetition Loops Lock? Truncation Sampling, Bounded Penalties, and a Blind Spot in DRY
Language models sometimes repeat verbatim until the generation limit. We study the exit rather than the entrance: given that a span has already been repeated $n$ times, when is escape still possible? Along the loop the context is deterministic, so loop survival under any sampler…
Tensor Network Datamodels for Training Data Attribution
Datamodels predict how a model's behavior changes when different parts of its training data are included or removed. However, linear datamodels assign fixed contributions to training units and cannot represent non-additive joint effects. We introduce Tensor Network Datamodels…
fxBench: Evaluating and Understanding Formula Suggestions in Spreadsheets
AI summaryThe paper introduces fxBench, a manually curated benchmark of 503 spreadsheet formula suggestion tasks, and evaluates various models to identify key bottlenecks in context gathering and formula generation. The best-performing configuration achieves only 58.8% accuracy, highlighting significant room for improvement in real-time spreadsheet assistants.
How Does Muon Work in Private Deep Learning? Scalable Differentially Private Training via Orthogonalization
AI summaryThis paper introduces DP-Muon, a differentially private variant of the Muon optimizer that uses orthogonalized updates. It proves that DP-Muon achieves standard privacy-utility convergence bounds without bounded gradient assumptions and matches or improves upon DP-SGD and DP-Adam empirically.
Gaussian Process Bandits with Paired Auxiliary Observations
AI summaryThis paper introduces a variance-reduction method for Gaussian process bandits that leverages low-cost auxiliary data alongside paired online observations. The approach provides theoretical regret guarantees and demonstrates consistent improvements in cumulative and simple regret over standard GP-UCB across benchmark tasks.
From Overparameterised Expressions to Parsimonious Models: Fisher-Geometric Pruning for Symbolic Regression
Genetic programming symbolic regression (GPSR) often returns expressions that are larger than necessary, with dormant branches and finely tuned algebraic configurations that fit sampling noise. Existing complexity-control strategies, such as multi-objective size-accuracy search…
Paired-Outcome Learning for Selective Spacecraft Attitude Capture under Torque Limits
AI summaryThis paper introduces a paired-outcome learning method to selectively modify spacecraft attitude controllers under torque limits. The approach reduces aggregate attitude error by 4.67% relative to a baseline without causing individual performance regressions.
Evidence Field: Adaptive Covering Programs for Medical Imaging
Medical image classifiers typically return predictions without a native contract for independently reconstructing their spatial decision computation. Evidence Field is a trainable spatial-program architecture that jointly produces class scores, selected covering programs and…
Global Convergence in Deep Networks via the Second Law of Thermodynamics
AI summaryThe paper analyzes continuous-time Langevin dynamics with noise for finite-width deep nets, proving that noise preserves Neural Tangent Kernel stability and yields a non-asymptotic training loss bound, guaranteeing convergence to a finite loss floor under explicit width and temperature conditions.
Process-Aware Event Log Augmentation for Process Prediction under Labeled Data Scarcity
AI summarySiamSA-PPM augments event logs with self‑supervised Siamese learning and control‑flow–aware transformations, creating realistic trace variants that improve prefix representations. It outperforms state‑of‑the‑art baselines on next‑activity and final‑outcome prediction in diverse real‑world logs.
Martingale Posterior Cox Processes
AI summaryMPCP estimates a Cox process intensity by treating unobserved events as latent variables, eliminating complex priors and enabling a GPU‑friendly copula‑based predictive update. It converges theoretically and runs far faster than existing Bayesian methods while matching or exceeding their performance.
Who Gets Unmasked? Sampler-Induced Bias in Diffusion Language Models
AI summaryDiffusion language models exhibit a sampler-induced bias where confidence-based token commitment rules amplify stereotypes and distort distributions. The authors propose Fork-First Sampling to resolve this bias by deciding critical choices before text generation.
Is User Calibration Still Necessary for Continuous Myoelectric Control? A Closed-Loop Comparison of Calibration-Free Controllers
The need for user-specific calibration remains an often-cited barrier to widespread adoption of myoelectric control. Cross-user models trained on large multi-user datasets can now recognize gestures from unseen users without calibration, but they have been evaluated almost…
Tilted Mondrian: Triage and Mitigation of Fine-grained Disparities in Conformal Prediction
AI summaryTilted Mondrian adjusts standard Mondrian conformal prediction to reduce fine‑stratum coverage and set‑size disparities while preserving group coverage guarantees, using propensity standardization to quantify severity and a method‑agnostic bound to measure achievable fairness gains.
First-Order LoRA Training with Phone NPUs
AI summaryFirst‑order LoRA training is ported to phone NPUs, executing all frozen‑weight multiplications on the NPU while keeping adapters and optimizer on CPU, and applying precision‑aware gradient guards; this yields on‑device fine‑tuning speeds comparable to GPU and 3–4× faster than CPU on Snapdragon.
Efficiency Measured in English: Tokeniser Coverage Bounds Small Language Models on Indic Scripts
Small Language Models with quantisation and parameter-efficient adaptation are widely offered as the route to multilingual AI under compute constraints, and their efficiency is almost always reported as a single memory figure. We show that this figure is measured in English and…
Pair-Aware Optimization: Coordinating Exploration and Refinement for Jailbreaking LLM Agents
AI summaryPair‑aware optimization coordinates zeroth‑order prefix search with first‑order suffix refinement, selecting the best prefix‑suffix pair. Experiments show the hybrid ZO‑FO strategy outperforms single‑optimizer attacks, and the optimization order significantly affects jailbreaking success.
Compose the Horizon: Parameter-Shared Blockwise Forecasting under a Sequential-Compute Trade-off
AI summaryRFD factorizes long-horizon prediction into a single compact, parameter‑shared operator that emits successive future blocks, trading a few hundred thousand parameters for sequential computation. On ten baselines it achieves lower error in 23/24 cells with 512–720 step context versus 96.
Recurrent Temporal Restricted Boltzmann Machines as Unsupervised Encoders for Biomechanical Movement Sequences
AI summaryRecurrent Temporal Restricted Boltzmann Machines encode windows of biomechanical motion into fixed‑length embeddings that capture temporal structure. On simulated reaching data, the embeddings recover task progression without labels; on real gait data they slightly beat PCA/autoencoders, but recurrence adds little benefit.
A Statistical Viewpoint on Adam, Shampoo and Muon
AI summaryThe paper proposes a statistical framework that whitens gradients using their covariance, leading to a diagonal method (MND) that matches Adam and outperforms Shampoo variants on GPT-2‑scale models, and a Muon variant (MN‑Muon) that improves stability and validation loss.
An Impossibility Result for ETF Target Geometry, and an Equinorm Remedy for Imbalanced Analytic Continual Learning
AI summaryThis paper shows that combining analytic continual learning with expanding ETF targets causes severe catastrophic forgetting. It instead proposes equinorm weight post-processing, which substantially improves tail-class and average accuracy under severe class imbalance.
LLM-Based Evaluation Planning for Heterogeneous Optimization
AI summaryREP-HO uses an LLM to plan which objectives and constraints to evaluate in heterogeneous optimization, replacing hand‑crafted metrics. It matches or outperforms MFE‑NSGA‑III across diverse regimes, adapting evaluation costs and re‑allocating budgets when surrogate reliability changes.
FlowMaker: From Tasks to Cost-Efficient Generative AI Workflow Deployments
AI summaryFlowMaker jointly optimizes workflow structure, model choices, and deployment configurations for generative AI tasks, using constraint-guided, load-aware pruning to make on‑the‑fly decisions. It boosts goodput by up to 10× and cuts cost by up to 2.8× over existing baselines.
Bellman Language Models: Solving the Recurring Decisions of LLM Agents
AI summaryBellman Language Models (BLMs) discretize an LLM agent’s conversation into states, estimate transition probabilities and costs, solve the resulting MDP once, and reuse the optimal policy, reducing model calls by up to 98% and raising task success from 29% to 96% on telecom benchmarks.
The Missing Compliance Class: LLM Graders That Collapse Jailbreak Success, and a Free Check
Jailbreak papers report an attack success rate, the share of responses that deliver the harmful request, and read it as a property of the target model. We hold the target model, the prompts and the responses fixed and change only the grader. Three dense graders of 8B and 9B…
RAFT: Redundancy-Aware Input Filtering for Temporal Graph Neural Network Training
AI summaryRAFT prunes redundant temporal events in TGNN training by tracking memory changes, filtering unnecessary prediction targets and historical context. It cuts inputs by ~35 % on average (up to 97 %) and training time by ~53 % with minimal precision loss.
Sample-Efficient Reinforcement Learning with General Utilities via Neural Dual Critics
AI summaryThe paper proposes a neural-actor critic algorithm that learns a dual pseudo-reward directly, avoiding occupancy estimation, and shows $\widetilde{O}(T^{-1/2})$ convergence and $\widetilde{O}(\epsilon^{-2})$ sample complexity for general concave RL utilities.
Measuring Data Diversity with Embeddings: A Taxonomy and Benchmark for Diversity Measures
Diversity is a critical property of text corpora, with direct implications for training data curation, test set design, and model generalization. However, its measurement remains fragmented and poorly understood. Existing embedding-based diversity measures vary widely in how…
Query-Dependent Reading for Quantized Inner-Product Estimation: A Minimax Separation
We study quantization for inner-product estimation when storage and reading have separate budgets. In the static bit-probe model [Yao, 1981], an encoder stores $w$ bits about $x$ before seeing a query $y$, and a reader probes at most $r<w$ bits to estimate $x^\top y$. Both…
Scalable and Adaptive Distribution Regression via Random Features
AI summaryRaDR uses random feature approximations to achieve scalable distribution regression with a finite-sample guarantee, matching the excess risk of exact kernel methods under a feature-budget constraint. A-RaDR adaptively assigns features via a score-guided scheme, improving accuracy while keeping the speed advantage.
LastWords: Boring Problems Revisited
AI summaryThe paper shows that modern frontier models (e.g., GPT‑4) have practically solved the number name reading subproblem of text normalization, but it argues that smaller, resource‑efficient methods could still address remaining edge cases without requiring massive models.
Distribution-Based Regularization for Reward Fine-Tuning Diffusion and Flow Models
AI summaryDistribution-Based Regularization (DFR) augments reward fine‑tuning of diffusion/flow models with a per‑sample, distribution‑dependent reward, directly penalizing drift from the pretrained model. It achieves the same peak reward as KL‑regularization while boosting diversity up to 29%.
A Symmetry-Matching Approach to Blind-Spot Reduction in Sparse Autoencoders
AI summaryAdds a differentiable Gini regulariser to sparse autoencoder training that penalises uneven perturbation responses, raising lower-tail sensitivity and reducing profile imbalance on GPT-2, Gemma 2, and Qwen 2.5 without hurting reconstruction, improving safety‑relevant feature alignment.
ATLAS: Expert-Guided Sustainable Satellite Backhaul for Off-Grid Wildlife Monitoring
Remote wildlife monitoring increasingly depends on high-fidelity cameras and sensors deployed far from grid power and terrestrial networks. Low Earth orbit satellite links can provide connectivity in these settings, but their high power draw and variable bandwidth, combined with…
Arctic-SWE: Revisiting the Long Horizon Agent Training by Modeling-System Co-Design
AI summaryArctic-SWE introduces an asynchronous reinforcement learning framework for long-horizon software engineering agents through model-system co-design. Using optimized infrastructure and curated data, it enables a 4B model to achieve 65.0% on SWE-bench Verified.
CARTA: Self-Supervised Reusable Transaction Embeddings via Heterogeneous Tokenization and Cross-Attention
AI summaryCARTA learns a 16‑dim, label‑free embedding of payment transactions using heterogeneous tokenization and InfoNCE. The frozen embedding boosts fraud detection, outperforming raw features at 1% labels and beating frequency encoding on large tabular data, while remaining reusable for downstream tasks.
DF-PALM: Dynamic Fusion Prompt Learning for Audio--Language Models
Few-shot audio classification requires audio--language models to distinguish variable acoustic events from limited labeled data. PALM's main formulation adapts encoded class representations without changing pretrained encoders, but its class representations remain…
Means-Ends Analysis Helps LLMs Plan in Continuous-State Domains
AI summaryMedaka uses Means‑Ends Analysis to guide LLMs in continuous‑state planning, recursively decomposing goals and tracking state. In Liquidworld, a liquid‑mixing benchmark, Medaka boosts accuracy across model families, showing that classical MEA techniques elicit better continuous‑state reasoning.
TrustAgent-RFF: Agentic Diagnostic Planning for Reliable RF Fingerprinting under Domain Shift
AI summaryTrustAgent‑RFF adds a frozen I/Q classifier with a diagnostic framework that measures counterfactual stability, prototype consistency, domain similarity, and signal quality to estimate prediction reliability without target labels, then uses an LLM planner to select diagnostics and a source‑calibrated decision layer to Accept, Reject, or Defer, reducing mis‑predictions under domain shift.
MLFont: Machine Learning-Driven Font Subsetting and Range-Compact Unicode-Range Slicing for Low-Latency Web Typography
AI summaryMLFont optimizes web font delivery for large character sets by using machine learning and genetic algorithms to group characters into efficient CSS unicode-range slices. This approach cuts CJK font loading latency by up to 75% and reduces bandwidth usage by over 80% without sacrificing glyph coverage.
T-MUON: Tensorizing Muon via t-SVD for Large Language Model Pre-training
AI summaryT‑MUON generalizes the Muon optimizer to tensors by grouping Transformer gradients into third‑order tensors and applying an M‑transform update; the resulting update is steepest‑descent under a spectral norm and converges. Experiments on GPT‑2 and LLaMA models show head‑wise T‑MUON with identity transform consistently lowers validation perplexity versus AdamW, Muon, and other transforms.
LowNet: Looped Wavelet Neural Networks with 24 Learnable Parameters for Solving Poisson Equations
AI summaryLowNet uses a 24‑parameter looped wavelet neural network that solves Poisson equations with high accuracy (up to 100.74 dB PSNR after three loops) and linear‑time complexity, making it compact and efficient for data‑specific applications.
Robust Direct Preference Optimization under Preference Label-Flip Attacks
AI summaryThe paper develops defense mechanisms for Direct Preference Optimization against label‑flip attacks, introducing Trim‑K and Soft‑Trim‑K methods that remove or downweight high‑impact pairs. It proves deterministic gradient‑perturbation bounds and shows these defenses achieve uniform policy deviation guarantees under bounded‑region assumptions.
Over-Defense in Coding Agents Begins at Silent Forks
AI summaryCoded language models over‑defend by taking unnecessary protective actions at silent forks—points where any action yields the same outcome. ForkDPO uses these forks as a learning signal to reduce such disproportionate behavior, cutting it from 70.3% to 58.9% on a 4B model.
ControlJEPA: Principled Trajectory Regularization via Lyapunov Tube Loss
AI summaryControlJEPA regularizes language model hidden-state trajectories using a Lyapunov contraction condition, penalizing deviations from a reference tube. It improves fine-tuning accuracy and data efficiency across six benchmarks without architectural changes.
CasePlay: Self-Play Reinforcement Learning from Case Reports for Medical Reasoning
AI summaryCasePlay uses document-grounded self‑play on medical case reports, generating questions from knowledge points and evaluating answers with rubric‑based quality feedback. It improves LLM medical reasoning by 3.1 overall points (6.3 on LiveClin) over prior self‑play baselines.
Revealing what benchmark scores hide: Jointly measuring the breadth and reliability of LLM capabilities using accessibility profiles
AI summaryThe paper defines “accessibility” as the probability a model succeeds across equivalent formulations of a question. It shows that many model pairs have similar average accuracy but diverge on Access@5 vs. Reliable@5, revealing a tradeoff between breadth and consistency of performance.
Label-Space Mismatch Distorts Cross-Cohort Evaluation in Breast Cancer Molecular Subtyping
AI summaryThe study shows that breast‑cancer subtyping classifiers fail cross‑cohort evaluation when training and test label spaces differ; adjusting the label set (label‑space compliance) improves accuracy by ~6–7 % and reveals that multimodal fusion adds no benefit over mRNA alone.
PRECOCIAL: Pretraining with Code Curriculums via Iterative Search for Accelerated Learning
Fully synthetic pretraining data generated by symbolic programs can bootstrap downstream training on naturalistic tasks, such as language modeling, but the generating programs have so far been designed by hand. We introduce PRECOCIAL, a method for automatically optimizing…
Certified Sparse Attention via Perturbed Retrieval
AI summaryAdds independent Exp(1) perturbations to KV scores, turning sparse key retrieval into probabilistic sampling that gives a Horvitz–Thompson estimate and simultaneous confidence bounds. The method reads only \(O(\varepsilon^{-2}\log(1/\delta))\) keys, achieves 4–8× lower 90th‑percentile error than baselines, and matches dense attention on benchmarks while speeding up decoding.
Block Recursive Transformers for Structured Parameter Sharing
AI summaryBlock Recursive Transformers partition the transformer stack into contiguous blocks and reuse each block recursively before progressing to the next, balancing local refinement and depth-wise transitions. Experiments on up to 1.7B decoder-only models show improved validation NLL and downstream accuracy over existing static sharing schemes.
SoK: PIAtlas: Decomposing the Unbounded Prompt Injection Attack Space
Prompt injection is the dominant security risk for large-language-model (LLM) applications and agents, yet the field still lacks a coherent map of how these attacks are constructed. More than twenty public datasets describe prompt injection using incompatible vocabularies, often…
UtoMe: Observation-Uncertainty-Guided Token Merging for Weather Foundation Models
AI summaryUtoMe introduces an observation-uncertainty-guided token merging framework for weather transformers that combines feature similarity with ensemble data assimilation spread. At a 50 percent token reduction, it significantly lowers forecasting error compared to standard token merging while maintaining reduced computational costs.
UpLoop: Upcycling Large Language Models with Gated Loop
AI summaryUpLoop is a post‑training technique that loops selected layers of a pretrained LLM with hidden‑wise gates, allowing each loop to refine outputs rather than overwrite them. Starting from Qwen3‑4B, UpLoop raises average benchmark accuracy by 6.17 % at R = 3.
No Escape from the Quadratic Barrier: Adaptive Attacks on Linear $\ell_2$ Sketches
Linear sketches estimate properties of high-dimensional inputs from a fixed low-dimensional projection. At constant accuracy, randomized Euclidean-norm sketches, notably the Johnson-Lindenstrauss transform, support exponentially many nonadaptive queries in sketch dimension $k$.…
Conjugate Compilation: Self-Improving Agentic Co-Compilation for Speculative Decoding
AI summaryThis paper introduces conjugate compilation, which jointly optimizes draft and target models for speculative decoding using an agentic search system guided by empirical execution feedback. This approach achieves a geometric-mean speedup of 1.37x over conventional execution across multiple GPU configurations.
Latent World Models Should Plan with Distances Along the Latent Manifold
AI summaryLatent world models misjudge distances in ambient latent space, causing planners to get trapped. By using a mutual‑nearest‑neighbor graph to compute manifold distances (NGC), the paper improves success rates to 95–100% on challenging tasks without retraining.
Multifidelity Monte Carlo Estimator for High-Dimensional Quantities of Interest
AI summaryThis paper extends multifidelity Monte Carlo estimators to vector-valued quantities of interest by using matrix-valued control-variate coefficients. The proposed method achieves significant variance reduction while providing closed-form optimal designs across a nested hierarchy of coefficient structures.
Decentralized Online Exp-Concave Optimization
AI summaryDecentralized Online Newton Step achieves logarithmic regret O(Nd(1−σ₂)⁻¹/² log²T) for exp‑concave losses, the first such bound for decentralized online learning. A nearly matching Ω(Nd(1−σ₂)⁻¹/² log(T/N)) lower bound confirms near optimality, and the method extends to online linear regression.
LLM Agents as Suppliers of Physical System Models: Calibration and Acceptance in Modelica
AI summaryLLM coding agents can generate Modelica models from client specifications and data for both simulated systems and a real magnet, but the handover leaves structural choices open. Acceptance is achieved by requiring the model to belong to a set of acceptable solutions rather than matching a single reference, and specifying residual thresholds improves pass rates.
Synergy-Aware Residual Flow Matching for Single-Cell Perturbation Prediction
AI summarySynGeneFlow predicts single‑cell responses to perturbations by learning a residual flow from an additive prior, incorporating gene‑interaction masks and deformation biases, and aligning distributions with MMD. It outperforms existing methods on genetic and drug perturbation benchmarks, especially for non‑additive synergy effects.
Integrable building blocks for symplectic neural networks: learned-type layers, fiber-preserving layers, and what a single layer can do
Exactly symplectic neural networks are built almost exclusively from shear maps and gain expressivity only through depth. We ask what else a single explicit symplectic layer can be. Observing that every layer in use is the exact flow of a Hamiltonian that is a function of…
Beyond the Teacher: Understanding and Preserving Distillation Overshoot in Agentic On-Policy Distillation
AI summaryThis paper identifies distillation overshoot, where student agents temporarily outperform their teacher before regressing. To prevent this, the proposed method masks conflicting teacher supervision on successful student actions, improving final performance by up to 15 percentage points.
Normalizing Flow for Neural-Enhanced Track-Before-Detect
Multiobject tracking (MOT) is a fundamental capability in autonomous navigation, radar and sonar systems, as well as 6G integrated sensing-and-communication (ISAC) networks. In this paper, we propose a neural-enhanced track-before-detect (TkBD) framework that operates directly…
PACT: NVFP4 Data-Flow LoRA Fine-Tuning on Consumer GPUs
AI summaryPACT accelerates LoRA fine‑tuning on consumer GPUs by using native NVFP4 GEMMs, reusing quantized RMSNorm outputs, and calibrating frozen weights with GPTQ, achieving lower memory usage and faster training with competitive accuracy on a 16 GB RTX 5080.
HCFM: Helmholtz Component Flow Matching for Time Series Anomaly Detection
AI summaryHCFM trains flow-based generative models for time‑series anomaly detection by splitting the vector field into divergence‑free transport, compression, and residual heads; this separation reduces Jacobian cost and improves detection accuracy on large univariate and multivariate benchmarks.
Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On
AI summaryThe paper shows that safety benchmark scores largely depend on the labeling policy used, not model capability. It demonstrates that minor changes in policy or a few parameters can shift scores across the full range, calling for explicit reporting of contamination and policy definitions.
The Optimal Sample Complexity of Pointwise Replicable Agnostic Learning
AI summaryDetermines the exact sample size needed for pointwise replicable agnostic learning on finite or countable domains: Θ(d/(ρ²α²)+log(1/β)/α²) examples for any binary class of VC‑dimension d≥1, matching lower bounds and removing previous log‑overheads.
Auto-MD-bench: Evaluating Autoresearch Agents for Real-World Medical Tasks
AI summaryAuto-MD-bench benchmarks 15 real-world medical tasks across 10 datasets, evaluating autonomous ML agents. The study finds that top agents match baseline primary metrics but have poorer risk calibration and robustness to population shifts, showing that single performance metrics are insufficient for clinical modeling.
CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow?
AI summaryCoEvo‑RSI uses a meta‑agent to inherit both harness designs and model checkpoints from a shared archive, aligning evolving execution scaffolds with learned behaviors to mitigate brain–body mismatch. This approach lets Qwen3.5‑9B, trained only with supervised fine‑tuning, outperform larger models on multi‑agent attribution, role‑aware text‑to‑SQL, and smart‑home tasks.
Conformal Prediction-Enabled Sequence-level Policy Optimization for Reasoning Language Models
AI summaryCSPO adaptively calibrates sequence-level clipping bounds via conformal prediction, targeting a fixed tail budget and adjusting for policy drift. It reduces miscoverage and gradient error, improving Qwen3-1.7B’s performance on mathematical reasoning and ALFWorld compared to GSPO.
Non-Acceleration of the Heavy-Ball Method in 1D
Despite having an asymptotically optimal rate of convergence on convex quadratics, the acceleration of Polyak's heavy-ball method and a characterization of its dynamics within more general settings have remained long standing open problems. Recently, the work of Goujaud et al.…
Self-organizing reinforcement learning: a swarm approaches the explore-exploit optimum
AI summaryA minimal foraging swarm balances exploration and exploitation via Levy flights and threshold cascades, reaching an ideal‑free distribution without learning or rewards. It attains ~65% of the oracle ceiling immediately, surpassing some RL methods early but not its final asymptote.
Compile Once, Run Cheap: Frozen, Portable Skills for Embodied Agents, and What Recursive Self-Improvement Adds
Embodied AI lacks the data-rich pretraining that gave language models their agents and skills, so new manipulation capability costs demonstrations per task or a frontier model in every episode. We study a third route: a strong general model acts only as an offline author that…
Polynomial Learning of Positive ReLU Networks under Gaussian Inputs
AI summaryPolynomial-time learning of one-hidden-layer ReLU networks with nonnegative outputs and Gaussian inputs: a width‑k target can be approximated to error ε using a nonnegative ReLU network of width at most k+min{k,d} with sample and time cost poly(d,k,ε⁻¹,log(1/δ)) and only bounded noise second moment.
Memory as Belief: Bayesian Evolution of Memory for Self-Evolving LLM Agents
AI summaryBEAM treats memory value as a context‑conditioned Bayesian belief, updating it with execution feedback. Across reasoning and code generation tasks, BEAM improves memory use and enables LLM agents to self‑evolve continually.
ClinSDT: LLM-Encoded Clinical Semantic Guidance for Sepsis Treatment via Decision Transformer
AI summaryClinSDT augments a Decision Transformer with LLM-encoded clinical semantics (Profile, Context, Advisory) and a treatment‑response forecaster to filter low‑uncertainty actions. On MIMIC‑III/IV it improves sepsis treatment policies over standard imitation, RL, and Transformer baselines.
Trackable Skill Library Lifecycle Management for Agentic Reinforcement Learning
AI summaryTrackRL uses online randomized controlled experiments within group-based policy optimization to isolate individual skill contributions in agentic reinforcement learning. This evidence-driven approach enables the principled lifecycle management of skill libraries, outperforming existing baselines in downstream evaluations.
Decision-Target Reversal under Informative Label Missingness
AI summaryThe paper shows that informative label missingness can improve classification under one loss but worsen it under another—“decision‑target reversal.” For binary Gaussian discrimination it derives critical cost ratios where this reversal occurs, extending to multiclass settings and demonstrating with simulations and a medical case.
Many Repairs, One Direction: Steering Code Models Toward Safer Cryptographic API Use
Cryptographic API misuse in large language model-generated code can compromise software security. While many existing approaches improve code security by modifying model parameters or inputs, how representation changes associated with different cryptographic API repairs can…
TIRE: Top-Down Formula Instantiation with Reliable Execution for Physics Problem Synthesis
AI summaryTIRE generates physics problems by instantiating reusable formulas and conditions, creating reference solutions before generation for reliable supervision. The resulting TIRE‑19k dataset improves fine‑tuned and RL‑trained LLMs by ~3 pp over comparable datasets.
HullProbe: Measuring Interventional Part Reliance in Maritime Vessel Classifiers
AI summaryHullProbe directly measures how much a maritime vessel classifier relies on each vessel part by swapping hulls or superstructures, using a Walsh transform to compute interventional influence. It shows that even after de‑biasing, about 46 % of the classifier’s decision is driven by the generic hull.
Spectral Adaptive Graph Transformer for Efficient and Scalable Traffic Forecasting
AI summaryThe paper proposes a Spectral Adaptive Graph Transformer that learns a directed, state‑conditioned spatial operator with linear sensor cost, combined with a sparse diffusion branch and frequency‑bin attention. It achieves state‑of‑the‑art accuracy on six benchmarks while using only a quarter of the memory and training twice as fast.
FHKAN: Fractional Hermite Kolmogorov-Arnold Networks for Data-Driven and Physics-Informed Learning
AI summaryFHKAN replaces KANs’ fixed scalar weights with learnable fractional‐order Hermite basis functions, eliminating saturation. On synthetic, image, and physics tasks, it achieves the lowest errors and highest accuracy across depths and problem types.
Learning without Overwriting: Structural Knowledge Fusion for Continual Learning
AI summaryKnowledge Fusion Network learns each new task in a lightweight specialist and fuses it into a global model using a novelty‑gated Structural Fusion Module. It achieves 71% accuracy on ten‑task Split CIFAR‑100 while adding only 25% more parameters, cutting forgetting by ~58% versus replay.
Indistinguishable Agents: Training Agent Privacy as Twin-World Indistinguishability
AI summaryIndistinguishable Agents trains agents to act the same in twin worlds differing only by a private fact, minimizing a distinguisher’s log-likelihood ratio; this guarantees Bayes-level privacy while preserving task performance on shopping and messaging tasks.
M-JEPA: Predictive Self-Supervised Learning for Molecular Graphs with Scaffold-Shift Evaluation on Tox21
AI summaryM-JEPA trains a molecular graph encoder with masked subgraph prediction and an EMA teacher, outperforming InfoNCE on ESOL (RMSE 2.12 vs 3.75) while running faster, and shows mixed but sometimes improved ranking and calibration on Tox21 scaffold splits.
World Models, Trees and Foundation Models for Retail Forecasting: Evidence from Four Public Data Sets
AI summaryThe paper benchmarks multiple forecasting models—world models, LightGBM, TFT, TimesFM, etc.—across four retail datasets at various grain levels (item, store, category, chain). It finds that an average of world model, LightGBM, and TimesFM performs best for most item/store and cross‑store tasks, while model choice varies for higher‑level aggregates.
An Information-Geometric Framework for OCS-Reconfigurable Interconnect Topology in Large-Scale AI Clusters
Optical Circuit Switches (OCS) make inter-pod wiring in AI clusters programmable, posing design questions on wiring choice, symmetry departure, and reconfigurations. We present a unified information-geometric framework to address these issues.A wiring is a point in a probability…
Dialogue Generation for Conversational Recommendation Using User Search Interactions and Knowledge Graphs
AI summarySearCl automatically generates 3,538 recommendation dialogues from search and click logs, using a probabilistic state machine and LLM agents guided by knowledge graphs. The generated conversations score higher than benchmark dialogues under an LLM-as-a-judge protocol.
Sampling and Curvature in First-Exit Comparisons for Momentum SGD
AI summaryThe paper analyzes how minibatch sampling and curvature affect the first exit time of momentum SGD in strongly convex problems, showing that even identical pointwise gradients can lead to different exit probabilities. It establishes an O(h) exit comparison for persistent sampling and provides empirical finite‑window exit brackets for various neural networks, achieving certified targets in most cases.
Shallow Probes Are Sufficient for Near-Oracle Personalization Depth Allocation in Federated Learning
AI summarySmall shallow probes (three rounds) capture almost all information needed to choose near‑oracle personalization depth in federated learning, achieving 98.3% oracle recovery and 0.0071 regret on CIFAR and PTB‑XL datasets, outperforming metadata and learned policies.
ChronoChart: Setup-Agnostic Channel Charting for Label-Free Indoor Localization
Accurate radio-based indoor localization remains challenging in multipath-rich and dynamic indoor environments. Supervised fingerprinting methods achieve high positioning accuracy by learning mappings from radio channel measurements to physical locations, but require costly…
RAIL-5G: A Challenging Real-World Dataset and Benchmark for AI-Based 5G Indoor Localization
Reliable radio-based indoor localization is essential for industrial automation, yet existing methods are often evaluated in environments that do not fully reflect the severe multipath and non-line-of-sight (NLOS) conditions encountered in real deployments. This limits our…
ViCoR: Estimating Visual Necessity via Counterfactual Residuals for Multimodal Medical Data Selection
AI summaryViCoR selects image-text pairs that truly need visual input by measuring counterfactual residuals and calibrating with medical supervision; on two backbones and multiple benchmarks, a 20% ViCoR‑selected subset outperforms full-data fine‑tuning and other selectors.
NeuroAct: Advancing EEG Analysis and Decoding via Dual-Loop Autonomous Research
AI summaryNeuroAct automates EEG research with a dual-loop system that tests hypotheses, selects interventions, and refines models using evidence from signal structure, representations, and training dynamics. It boosts macro‑average accuracy by 17.75 points over BrainAgent and improves balanced accuracy and correlation across four EEG tasks.
Calibration-Free Early Stopping for Chain-of-Thought Reasoning via E-processes
AI summaryCalibration-free early stopping for chain‑of‑thought LLM reasoning uses two e‑processes to sequentially accumulate evidence, guaranteeing a user‑specified false‑alarm rate without calibration. The method matches or outperforms calibration‑based policies across models and datasets.
LiftDiff: Guided Discrete Diffusion for Quantum Code Construction
The realization of fault-tolerant quantum computation relies on Quantum Error-Correcting Codes (QECC). Of particular interest are codes that combine a high encoding rate with sufficient distance. Existing algebraic constructions, such as lifted product (LP) codes, build a large…
DROD: Dual-Ratio Ordered Dropout for Federated Continual Learning on Heterogeneous Devices
Federated learning for capacity-heterogeneous clients and federated continual learning are each well studied on their own, but their intersection -- clients holding differently sized models while every client's data stream also drifts over time -- remains underexplored. We study…
Assessing the Empirical Evidence for Loss of Control from Agentic General-Purpose AI
Theoretical literature warns that advanced generalist agentic systems might pursue instrumental goals that are misaligned with human preferences, thus posing risks of irreversible loss of human control. The EU's Code of Practice for General-Purpose AI lists loss of control as a…
Estimating Causal Bounds from Observational Data With Neural Generative Models
Identifying causal bounds from discrete observational data is a fundamental problem in causal inference. While exact analytical bounds exist for certain graphs, this is not generally the case. In this paper, we introduce a framework for estimating such bounds using neural…
Improving the Expressive Power of Hybrid Sequence Models via Multi-Hop Retrieval
AI summaryHybrid models that combine few full‑attention layers with many efficient layers lack the ability to use the KV cache fully; by adding multi‑hop retrieval (MuRA), a single fixed‑size block can compute m‑function composition with m hops, proving a strict expressivity hierarchy and outperforming prior hybrids on synthetic tasks.
Persistent Representational Misalignment Induces Vision-Language Model Failures
Vision-language models (VLMs) fail at some tasks that are simple for humans---but why? We hypothesize that many such failures stem from VLMs' difficulties translating between vision and language representations, a misalignment the Platonic Representation Hypothesis predicts…
BatchParScale: Sparse Batch-Mixing for Sequence Models
AI summaryBatchParScale adds a sparse communication attention to parallel reasoning streams, enabling them to share information at selected positions. It improves language‑model loss over ParScale, outperforms unmasked baselines in masked pretraining, and works with LoRA fine‑tuning, but retrieval gains remain uncertain.
Emergent Sparsity: How Many Effective Variables Survive Coarse-Graining in AlphaEarth Embeddings?
AlphaEarth encodes each 10 m land pixel as a 64-component vector intended to support diverse global mapping tasks. Yet many scientific applications work at coarser spatial supports, for which fewer than 64 effective variables may suffice. We ask whether these vectors can be…
EasyResearch: Honest Regression, Forecasting and Signal Classification for Researchers Who Do Not Program
EasyClassifier showed that researchers who do not program can obtain honest classification results when the correct evaluation is the default. Most applied studies, however, predict a number, forecast a series or classify recordings, and each task has its own way of producing…
Federated Learning for Malware Image Classification under Data Heterogeneity
Malware image classification (MIC), in which binary files are converted into visual representations and deep neural networks are trained to identify malware families, has recently emerged as an effective approach for malware detection. Federated learning (FL) enables malware…
EG-GRPO: Evidence-Guided Step-level Credit Assignment for Deep Search Agents
LLM-based search agents demonstrate strong capabilities in long-horizon information seeking and reasoning, typically trained via reinforcement learning. However, existing approaches predominantly rely on trajectory-level outcome rewards, providing limited supervision for complex…
Bridging Concepts: Multi-Agent Reasoning over Scientific Concept Graphs for High-Impact Research Prediction
AI summaryThe paper builds scientific concept graphs from conference papers and proposes MiroGraph, a multi‑agent system that designs, extracts, and classifies features to forecast high‑impact concept pairs. MiroGraph outperforms existing impact‑forecasting and temporal link‑prediction baselines.
Data-driven forecasting of tropical cloud forest phenology: A dynamical systems and machine learning hybrid approach
AI summaryThis paper combines dynamical systems theory with Gaussian process regression to forecast tropical cloud forest flowering and fruiting cycles. The hybrid approach achieves higher predictive accuracy for ecological phenology using a 35-year observational dataset.
A discrete-continuous-time stochastic framework for attention-deficit/hyperactivity disorder-related symptom modeling and pharmacological intervention via methylphenidate
AI summaryThe paper develops a discrete‑continuous stochastic control model for ADHD, treating methylphenidate as an optimal policy that redirects neural trajectories from pathological to normal states. It couples the model with a hybrid CNN‑ViT that infers latent states from fMRI data.
The Explainability Illusion: Must ViTs Exhibit Patch-Decoding Interpretability?
Transformers are widely assumed to exhibit token-decoding explainability. In particular, in vision transformer (ViT) classifiers, decoding individual patch tokens has become a common diagnostic tool due to its tendency to yield human-interpretable outputs. In light of the…
World Model Interpretability: From Decoding Representations to Understanding Latent Worlds
Emerging Joint-Embedding Predictive Architectures (JEPAs) are moving beyond representation learning towards latent world models that support prediction, planning, and control. Yet their latent worlds naturally lack a human-interpretable interface. In this Perspective, we argue…
Learning When to Abort in Transformer Planning: A Mechanistic Study
Planning does not by itself imply the ability to recognise errors, but recognising errors can improve planning capacity. In this paper, we study this phenomenon in Transformer planning. To analyze the inner mechanism, we follow the approach of Wang et al. (2024) and abstract…
Where and Why Search-Augmented LLMs Fail When Reasoning over Rapidly Evolving Knowledge
AI summaryTREK is a framework and benchmark that diagnoses why large language models struggle to reason about fast‑changing facts. It shows that the main failure lies in post‑retrieval reasoning, where models get misled by conflicting evidence rather than retrieval errors.
Where Groups Split: A Probe Prior for Nonzero Within-Group Variance in RL Post-Training
AI summaryThe paper introduces Aperture, a probe‑prior that estimates the probability a training group is unanimous or split using short rollouts. Applying this static prior improves greedy accuracy and increases nonzero advantage groups compared to uniform GRPO, even when combined with dynamic sampling.
Follow the Mixed Groups: Ridge-Tracking Gaussian Sampling for Zero-Variance Web Rollouts
AI summaryRIDGE modifies Gaussian curricula for group relative policy optimization by dynamically adjusting the Gaussian’s center and width using Beta posteriors of prompt groups, weighting samples by marginal likelihood and cost. It improves math QA accuracy by ~3–4 points while reducing rollouts and tool calls by ~30%.
A Distribution-Free Separation Between Learned and Fixed Features
AI summaryA two‑layer ReLU network learns a constructed hypothesis class under any input distribution within any specified runtime, while any linear predictor with fixed features would require super‑polynomial dimension, showing a distribution‑free separation between learned and fixed features.
Formally Verified Autoresearch for Analyzing the Expressivity of Multi-Head Attention
Multi-head attention is a central component of the transformer architecture, yet how its expressivity depends on the number of heads remains poorly understood. We study a one-layer, attention-only transformer with additive positional embeddings and define the head complexity…
Predicted Class Correction: Conformal Thresholds for Classes with No Calibration Data
Class-conditional conformal prediction calibrates a separate score threshold for every class and therefore requires labelled calibration examples of every class. In long-tailed deployments these are exactly what is missing: rare classes are left with few or no held-out examples,…
CHSM8: A Chess Transformer that Follows Natural-Language Style Prompts
We present CHSM8, a 210M-parameter chess transformer that plays moves conditioned on a free-text description of a playing style, e.g. "go for the Sicilian Defense", "solid, safe, risk-free chess" or "You are Magnus Carlsen in his prime". The model is pretrained without search on…
CD-OPD: Contextual Digest for On-Policy Distillation
Teacher takeover in on-policy distillation (OPD) can produce incorrect responses even when the student and teacher independently answer the same problem correctly, a failure we quantify using Collaboration Failure Rate (CFR). To address it, we propose Contextual Digest for…
Operator-Mismatch Contrastive Learning for World-Relative Trajectory Coherence
AI summaryThis paper introduces Operator-Mismatch Contrastive learning to train world models to distinguish physically valid futures from those governed by wrong latent rules. The method successfully resolves operator-blindness, where standard models score high on generic plausibility while failing to track specific world dynamics.
GreenNIDS: A Benchmark Workflow and Analysis Dataset for NetFlow Intrusion Detection Experiments
Comparing the predictive performance and computational cost of flow-based intrusion detection experiments requires both quantities to be available for each run. GreenNIDS connects benchmark execution on public NetFlow datasets to a reusable analysis dataset. Model…
Multi-site Learning for Forecasting Day-Ahead Wind-Power Production for Onshore Wind Farms in Great Britain with Deep Neural Networks
Accurate day-ahead wind-power production forecasting is increasingly important for energy system operations, market activity and balancing Great Britain's grid as wind generation grows. In this study, we train multi-site deep neural networks (DNNs) using pooled historical…
Whistle and Wrong Lyrics: Synthetic Queries Are Not Enough for Real-World Query-by-Melody
People look up a song by humming it, but also by whistling it or singing it with the wrong words. Published query-by-humming systems and benchmarks focus on hums and sung queries, and no public dataset pairs whistles or wrong-lyrics singing with song recordings. We study both…
Mining the Self-Rewriting System: A Vision for Data Mining on AI-Native Infrastructure
Classical data mining treats data as generated outside the learner. Emerging AI-native infrastructure invalidates this assumption: as agents observe system state, act, update retrieval stores and models, and promote new artifacts into the serving path, their own decisions…
TILDE: Teacher-in-the-Loop Distillation from Embodied Foundation Models
Large-scale action pretraining equips Embodied Foundation Models (EFMs) with strong manipulation capabilities, with particular strengths in generalization and closed-loop robustness, but reproducing such pretraining requires substantial robot data and computation. To transfer…
Simulating Infant Senses with Age-Tuned Networks
What a newborn sees, hears and feels, and how this changes month by month, has been measured sense by sense, with different methods and in different units. We bring these measurements together for the first two years of life and ask whether small neural networks tuned to a given…
Outcome-Anchored Credit Reallocation for Agentic Knowledge Graph Question Answering
AI summaryThe paper introduces Outcome-Anchored Credit Reallocation (OACR) to assign fine-grained operation credits from final answer rewards in agentic knowledge graph question answering. OACR improves multi-turn reasoning performance across three benchmarks while reducing inference-time LLM calls and token usage.
Structured Relevance Determination and Channel Additive Gaussian Processes for Image Data
AI summaryStructured RBF kernels tie lengthscales across adjacent pixels and colour channels, reducing hyperparameters for image inputs. Combined with channel‑additive GP decomposition, they achieve better predictive performance than isotropic or fully anisotropic baselines on MNIST and satellite image datasets.
The Hidden Look-Ahead: Point-in-Time Leakage in Electricity Market Data
Empirical studies of electricity prices routinely state that their models use only information available at the time of the trading decision. For weather-driven forecasts drawn from commercial archives, this condition is harder to meet than it appears, and whether it holds can…
Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting
Explicitly modeling periodicity has proven to be an efficient approach for time series forecasting, where the residual, obtained after removing the periodic component from the time series, is typically treated as unstructured noise. However, in this work, we show that such…
MOSAIC: Multi-Objective Search for Allocating Inference Cache Across LLM Decoder Layers
The Key-Value (KV) cache is the main memory bottleneck of long-context LLM inference. Token eviction methods such as SnapKV, H2O, AdaKV and L2Norm shrink it, but they give every transformer layer the same budget, although layers differ in which tokens they need. We present…
RARE: Reinforcement Learning with Adaptive Rubric Evolution for Open-Ended Generation
Verifiable rewards have advanced reinforcement learning in mathematics and code. However, many valuable open-ended tasks share three properties: no automatic verifier of overall quality, multiple requirements to satisfy jointly, and multiple valid responses to each query.…
Enhanced Masked Diffusion Unlearning
Masked diffusion language models (MDLMs) generate text by iteratively unmasking tokens with bidirectional context. As these models reach real-world deployment, machine unlearning becomes essential. This process aims to remove the influence of a designated "forget set" without…
Semi-supervised learning on singular data sets
AI summaryWe model data clouds with subanalytic sets and apply a Morrey-type embedding for Sobolev spaces on possibly singular measures. A greedy Monte Carlo scheme learns classifiers on point clouds from algebraic varieties and 7×7 MNIST, achieving competitive accuracy.
BiDex-100: A Comprehensive Benchmark for Bimanual Dexterous Manipulation Across 100+ Simulation Tasks
Bimanual manipulation with dexterous hands has attracted growing attention in research on robotic action models. However, few studies have developed benchmarks specifically tailored for dexterous manipulation, leaving the dexterous manipulation capabilities of existing models…
Reliability-Preserving Compression for Deployable Medical Image Segmentation
Post-training compression of pretrained medical segmentation models commonly uses quantization, pruning, or distillation to reduce model size and computational cost. Existing medical segmentation quantization studies mainly evaluate Dice, model size, memory, and latency, but…
On Reasoning Correctness Directions
AI summaryThe study trains a 3‑million‑parameter model on ChainSearch, an arithmetic reasoning task, and uses linear probes to find internal directions that predict reasoning correctness. These directions are statistically significant but do not causally improve reasoning, challenging the linear representation hypothesis.
Near-Optimal Last-Iterate Convergence of SGD in the Smooth Interpolation Regime
We study the last iterate of stochastic gradient descent (SGD) with a constant stepsize on smooth convex objectives in the low-noise and interpolation regimes, where the variance $\sigma_\star^2$ of the stochastic gradients at the optimum is small or zero. We establish that…
Consistent LLM Call Handling Across Java Frameworks: A Reference Implementation and Benchmark
Java applications reach large language models (LLMs) through an integration framework, and rely on it to handle a failing provider: to retry what can be retried and nothing else, to wait as long as the provider asks, never to present a truncated answer as complete, and to record…
Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models
AI summaryProbabilistic limit order book forecasting is improved by LOBGrad, a conditional diffusion model with an autoregressive RNN encoder, which yields better queue metrics, calibrated intervals, and theoretical excess‑risk guarantees compared to encoder‑free baselines.
Steered Not Stirred: Effective Reward Tilting using Surrogate Reward Mixing
Continuous-time generative models, from diffusion models, flow matching, and now flow maps, owe their practical value to high-fidelity generation and steerability. In many applications, a ground-truth latent ``oracle'' reward $R^*$ is unavailable or expensive to evaluate, e.g.,…
From Seeing to Reporting: A Theory of Reliable Visual Set Generation in Vision-Language Models
Vision-language models (VLMs) are asked to report what they see: to count objects, box them, parse a screen or pick a click target (GUI grounding). We argue that reporting is part of the visual task, not merely its formatting. In a dense scene with 173 shapes, GPT-6.1 lists 173…
KPI-Aware Multimodal Anomaly Detection for 5G and Open RAN Telemetry
Reliable anomaly detection is a deployment bottleneck for 5G/Open RAN observability and emerging self-optimizing 6G systems, where hundreds of heterogeneous KPIs, severe class imbalance, and strict alert budgets make purely numerical scores brittle. We present KAC, a…
Dynamics-Aligned Representation Learning under Induced Partial Observability
Model-based reinforcement learning (MBRL) has enabled sample-efficient learning of continuous control policies from high-dimensional observations through learned latent dynamics models. These models are typically trained with observation reconstruction objectives, which provide…
ATEST: Adaptive Resource-Aware Test-Time Scaling for LLM Inference
Test-time scaling (TTS) improves large language model (LLM) reasoning by sampling multiple answers, but fixed sampling budgets ignore both query difficulty and the changing energy cost of inference under GPU contention. We measure up to a 4.1-fold increase in energy for the same…
FracTS: Hierarchical and Autoregressive Time Series Generation
Time series are ubiquitous across different domains, and the growing use of synthetic data for privacy-preserving sharing, data augmentation, and stress testing makes time series generation increasingly important. Early generative approaches are predominantly autoregressive,…
Breaking the 1/3 Barrier for Approximate Nash Equilibria in Symmetric Bimatrix Games
We give the first unconditional family of polynomial-time algorithms that breaks the $1/3$ approximation barrier for symmetric bimatrix games. For every fixed $\delta>0$, an algorithm in this family computes a symmetric $(1-1/\sqrt{2}+\delta)$-approximate Nash equilibrium,…
From Clinical Risk Prediction to Operational Decisions: An Uncertainty-Aware Framework for ICU Staffing
Advances in electronic health records (EHR) and large-scale critical care databases have enabled the development of machine learning models for the early detection of patient deterioration. While these models often achieve strong predictive performance for outcomes such as…
Reinforcement learning on the EPRI AI-ccelerating Unit Commitment benchmark: diagnosing over-commitment and guiding MILP with learned policies
Reinforcement learning (RL) is increasingly proposed as a fast solver for unit commitment (UC), but standalone agents remain far more expensive than mixed-integer linear programming (MILP). In the first study of RL on the 51-generator, 72-hour benchmark of the Electric Power…
Beyond the record: evaluating probabilistic AI weather forecasts deep in the tail
Can probabilistic artificial intelligence (AI) weather forecasts retain skill beyond their training domain? We compare four state-of-the-art AI models (ArchesWeatherGen, U-Cast, WeatherNext2 and GenCast-mini) with the physics-based IFS-ENS over land in 2022, at 1.5{\deg}…
Narrow Decision Models: A System-1 Architecture for Calibrated, Typed Decisions
System One models, introduced with TypeSafe's Jev, answer typed questions about a state with calibrated probabilities instead of generated text: choosing among options described in natural language (Choice), placing the state on an ordered scale (Score), or judging a yes/no…
FTFM: Factorized Tabular Foundation Models
Tabular foundation models (TFMs) are becoming a powerful alternative to conventional tabular learning, adapting to new datasets in context without task-specific training. Current architectures, however, face a trade-off: models that preserve cell-level representations are…
Concurrent two-scale topology optimization of printable bone scaffolds via rectified-flow generation and exact latent-space refinement
Bone scaffolds must reconcile a macroscale mechanical response with microscale geometric feasibility, yet conventional approaches to their design restrict attainable constitutive properties. This work presents a concurrent two-scale topology optimization framework; a conditional…
Benchmark for Structured Retrieval-Augmented Generation in Perovskite Photovoltaics
While retrieval-augmented language models increasingly serve as literature assistants in the physical sciences, existing benchmarks evaluate them primarily on qualitative text overlap or through model-based scoring rather than against verified experimental ground truth. In…
Finite Interaction Lens: Structured Precision Sharing for Vision-Language Models
Identifying which visual interventions interact in a vision-language model requires comparing their joint and separate effects. The resulting four-output difference can be small enough for rounding error to obscure it. We introduce the Finite Interaction Lens (FIL), a…
CVTA: Cross-Variable Temporal Attention for Multivariate Irregular Time Series Prediction
Irregularly sampled multivariate time series arise in many real-world domains such as healthcare, climate science, and large-scale monitoring systems. Characterized by asynchronous observations and heterogeneity across variables, such time series pose significant challenges for…
Divergence based Ranked Preference Alignment for LLMs
Alignment of large language models typically relies on pairwise preference feedback, though ranked feedback can capture more comprehensive information across multiple responses within a single annotation. However, ranked-preference alignment is underexplored in practice compared…
Distributionally-Robust Policy Learning with Observational Data
Policy learning from observational data is fundamental to applications ranging from personalized medicine to welfare-program targeting, but standard approaches optimize empirical performance and may fail when the deployment distribution differs from the training distribution. We…
Cognitive Map in Large Language Models
AI summaryLarge language models build abstract relational structures that act as cognitive maps, inferring unseen relations and generalizing to synthetic entities. The study shows models store pairwise links, bind entities via sparse key–value pairs, and can be manipulated to implant false memories.
Text Dominance in Paralinguistic Conflict: Tracing Suppression Through Audio LLM Circuits
Audio LLMs deployed in settings such as crisis hotlines and customer support are only useful beyond speech-to-text pipelines if they use acoustic cues rather than the transcript alone. We test whether they prioritise transcript content over paralinguistic cues (emotion, speaker…
Adversarial Post-Training of Foundation Models for High-Frequency Trading
While autoregressive foundation models trained on limit order book (LOB) data show state-of-the-art performance for market simulation, pre-training with cross-entropy neither penalizes the compounding of sampling errors over free-running rollouts nor targets desired market…
PRISM: Prior Relational Information for Self-supervised Modeling to Enhance Solubility OOD Generalization
Solubility depends on intermolecular interactions between solutes and solvents, and is central to drug discovery, chemical and materials process engineering. Its prediction requires out-of-distribution (OOD) generalization to unseen solutes, solvents, and solvent mixtures, where…
Safety Offset Learning: When Benign Fine-Tuning Misaligns LLMs
Safety risks from fine-tuning are commonly characterized by the harmfulness of the training data:harmful fine-tuning is considered dangerous, while benign fine-tuning is often assumed to preservealignment. We argue that this view misses a more fundamental factor-the behavioral…
Curriculum Learning Teaches LLMs Long-Horizon Implicit Reasoning, Up To A Limit
LLMs "reason" by externalizing intermediate steps as tokens through chain-of-thought (CoT), yet even frontier reasoning models hit abrupt performance cliffs on long-horizon tasks. Implicit reasoning (reasoning within a single forward pass) is useful during CoT generation, as the…
CheXFuse: Guided Differential Joint Attention Fusion for Robust Multimodal Chest X-Ray Classification in Data-Scarce Medical Settings
A central challenge in multimodal learning is how to fuse modalities that differ substantially in predictive strength. When one modality provides a much stronger representation than another, indiscriminate cross-modal interaction can introduce weak, redundant, or unstable…
Weighted State Souping for Efficient Long-Context Retrieval
State space models (SSMs) combine fast inference and fixed state size with competitive language-modeling performance. Recently, state souping has been proposed as a procedure whereby an SSM independently encodes chunks of tokens like documents, tool definitions, or memory…
Zephyrus-R1: Agentic Reinforcement Learning for Weather Reporting
AI summaryZephyrus‑R1 trains vision‑language agents to generate weather forecast reports by rewarding legibility and factual accuracy with a rubric plus entailment against NWS reports. After curriculum pre‑training, Qwen3.5‑4B improves six judge‑metric scores by 26%, matching competitive models.
Conditional Average Treatment Effect Distillation for Hypothetical State-Intervention Targeting Under Budget Constraints
Personalized public-health targeting must capture heterogeneous intervention benefits while remaining interpretable under resource constraints. Flexible causal-effect models can be hard to operationalize, while coarse subgroup policies may lose individualized effect information.…
Clean Forcing: Drift-Resistant Autoregressive Video Diffusion with a Frozen Base
Autoregressive (AR) video diffusion models generate streaming video of arbitrary length by conditioning each new chunk of frames on those already generated. However, they suffer from exposure bias, where a model trained on clean context must continue from its own imperfect…
Uncertainty-Gated SHAP-Dirichlet Refinement for MRI-Based Depression Classification
Structural MRI-based classification of major depressive disorder (MDD) remains challenging due to subtle, heterogeneous anatomical changes. We propose uncertainty-gated SHAP-Dirichlet refinement, a post-hoc reliability layer that improves predictions without retraining the…
Sampling of Signals on Multigraphs: From Removable and Uniqueness sets to Spectral Proxies
In this paper, we study the sampling of bandlimited signals on multigraphs, where several graphs with different shift operators are defined on the same set of nodes. Since the shift operators do not commute in general, they admit only a joint block diagonalization, and the…
Spherical Fixed-Point Reasoners: Adaptive Loop Halting on the Hypersphere
Looped (weight-tied) transformers enable adaptive computation by allocating more iterations to harder inputs. Fixed-Point Reasoners (FPRM) achieve this with a damped fixed-point solver, but operate in an unbounded Euclidean residual space with coupled rescaling and relative-norm…
Consistency Training Suppresses Cue Following Without Teaching Cue Selectivity
Language models often change answers because of a cue, a hint such as "I believe the answer is B" that gives no valid evidence. Consistency training (CT) trains a model to answer the same with and without the cue. Four CT methods apply this idea to output tokens (BCT), residual…
OPDArena: Understanding Learning and Forgetting in On-Policy Distillation
AI summaryOn-policy distillation’s effect on learning and forgetting is studied with OPDArena, comparing SFT, RLVR, and OPD across five models and teachers. RLVR teachers improve target tasks while preserving prior skills, whereas SFT teachers cause more forgetting, especially for out‑of‑domain changes.
A "feature ODE" describing the learning behavior of shallow MLPs on simple functions
AI summaryThe paper introduces a “feature ODE” that approximates the gradient‑flow dynamics of two‑layer MLPs on anisotropic Gaussian data, showing that trained networks on simple targets (Hermite or staircase functions) largely follow the ODE’s loss trajectory, an unexpected but useful correspondence.
Knowledge Manifold Optimization for Scientific Discovery
The rapid expansion and specialization of scientific literature make it increasingly difficult to identify not only what is known, but where new and still supportable knowledge may emerge. We introduce Knowledge Manifold Optimization (KMO), a discovery-oriented extension of the…
Can Metadata Fix the Gauge? Calibration Turns Sparse Multi-Source Learning from Sparse PCA into Sparse Mean Recovery
Source labels tell us where a sample came from, but not how that source oriented its coordinates. In a sparse Gaussian multi-source model with source-specific sign gauges, that distinction changes the learning problem. Without calibration, pooling across sources erases the mean…
Arbitrage-Free Implied Volatility Surface Reconstruction via Additive Martingales
AI summaryAddMart reconstructs a full implied volatility surface from few option quotes by learning non‑negative increment densities of an exponential martingale, guaranteeing static no‑arbitrage by design. It achieves competitive accuracy on real data while enabling VaR, CVaR, and crash probability estimates.
LOOP: Recovery-Aligned Iterative Layer Pruning via On-Policy Distillation
Layer-level pruning is attractive for large language model (LLM) deployment as it reduces the model size while preserving a dense architecture that can be efficiently executed on standard hardware. Current methods typically follow a one-shot paradigm, which first removes…
MT-CC: Multi-Group Temperature Scaling for Asymmetric Calibration Behavior in Class-Incremental Learning
As deep neural networks are required to continuously adapt to evolving data, class-incremental learning (CIL) has become an active research area. However, most existing works primarily focus on improving accuracy while overlooking confidence calibration, which measures the…
Progress-Guided Preference Generation for Aligning Generative Control Policies
Generative policies have achieved strong performance in imitation learning tasks, but their success is often constrained by the limited availability of expert demonstrations. Although preference-based fine-tuning can leverage suboptimal rollouts to improve performance, existing…
Stochastic Tracking of Drifting Affine Fixed Points
Linear stochastic approximation (SA) is widely used to estimate fixed points from noisy observations, but most finite-time analyses assume that the underlying operator remains fixed. We study linear SA with time-varying affine operators, where the fixed points change over time…
Predictive Bandit Curriculum Reinforcement Learning via Gradient-Based Sample Group Discovery
Curriculum learning, which organizes training samples according to their learning priority, has been widely adopted in reinforcement learning to improve training efficiency. However, existing methods commonly construct curriculum units using predefined data properties and…
Evidence Sufficiency and Model Comparisons in MexRegQA: An Item-Level Audit of Mexican-Spanish Regulatory QA
Similarity-gated coverage does not establish that retrieved evidence contains the facts needed to answer a question. We audit Mexican-Spanish financial-regulatory QA and find that 35 of 135 covered items lack some or all load-bearing facts. On the post-hoc full-evidence slice…
Corrected Block Sparse Spectral Optimization: Exact Models, Finite Identification, and Local Dynamics
The block generalized-power method pairs threshold-based support selection with inexpensive Stiefel updates, but the standard block formulation conflates two nonequivalent objectives and leaves inactive components, threshold ties, and post-selection refitting ill-specified. We…
Learning to Couple: Context-Adaptive Joint Denoising for Diffusion Language Models
Diffusion language models accelerate generation by predicting multiple tokens in parallel, but factorized denoisers treat simultaneous predictions as conditionally independent. Decoding heuristics reduce the resulting errors by controlling which tokens are committed, but leave…
Anatomy of Off-Policy Policy Gradient: Importance Sampling, KL Regularization, and Baselines
Reinforcement learning has become a central tool for large language model (LLM) post-training, where policy gradient methods are routinely deployed off-policy, even though vanilla policy gradient assumes on-policy sampling. We study when such off-policy deployment is justified,…
Neural Video Coding under Limited Delay: Theoretical Optimality and System Design
AI summaryThe paper shows that a limited delay of K frames suffices to achieve joint‑coding rate–distortion optimality for K‑order Markov sources, and proposes Graph‑Factored Sequential Coding (GFSC) that augments existing causal codecs to exploit future context, improving RD performance and matching bidirectional codecs under the same delay budget.
A Calibrated Instrument for Detecting Disposition Drift in Language Models
Before language models are released, they are tested for harmful traits such as willingness to carry out dangerous requests, deception, power seeking, and sycophancy. The safe behavior these tests certify may degrade as models are modified, by topical fine-tuning, new training…
How Much Does One Wrong Answer Cost? Error Propagation in Constraint Pruning with Noisy LLM Oracles
Constraint pruning saves queries to a large language model (LLM) or another noisy oracle by skipping each yes/no question implied by answers accepted earlier, for example through transitivity. Entity-resolution systems and LLM-based rankers use this idea. However, their analyses…
Fundamental Limits of Information-Theoretic Secure Aggregation with User Dropouts in Federated Learning
Secure aggregation enables a server to compute the sum of users' gradients or model updates without revealing individual values. With user dropouts, correlated random keys protect local gradients while allowing recovery of the desired gradient sum. We consider three phases: key…
Bayesian Model Discovery with Language Models: Closing the Loop between Hypotheses and Experiments
Scientific model discovery requires proposing candidate models of an unknown system and designing experiments that discriminate among them. However, LLM-based approaches typically rely on in-context learning to predict outcomes and select experiments, lacking explicit model…
Cognitively Decoupled Meta-Induction for Out-of-Distribution Generalization
In-context learning (ICL) tackles downstream tasks conditioned on demonstrations. This paradigm relies heavily on neighbor-based matching between queries and demonstrations, limiting out-of-distribution (OOD) performance. For better generalization, rule-based inductive reasoning…
NADS: Navigator-Guided Data Selection for Mitigating Catastrophic Forgetting in Fine-Tuning
Catastrophic forgetting is a central obstacle in single-stage fine-tuning of large language models (LLMs) on a new task, especially when the original pretraining data are unavailable. Synthetic data offers a practical mitigation path, but candidate pools are often noisy and…
Which Answers Agree? Evaluating Self-Consistency from Incomplete Logs
Self-consistency answers a question by generating multiple responses and selecting the most frequent answer. Its evaluation requires both correctness labels and information about which responses agree; correctness labels alone generally determine neither voting accuracy nor the…
Remember Less, Adapt Longer: Bounded-State In-Context Reinforcement Learning with Recurrent Summary Memory
AI summaryRecurrent Summary Memory (RSM) is a Transformer policy that keeps a fixed-size working buffer and a compressed summary of past experience, allowing in-context reinforcement learning with constant memory and per-step cost. RSM sustains adaptation over eight times the training horizon using only 1 % of full-history memory, outperforming sliding-window, full-history, and other recurrent baselines.
Computation, Not Memory: Chain-of-Thought and State Tracking in LLM Agents
Language-model agents often reason before they act, and the value of monitoring that reasoning depends on how it relates to their actions. We ask how an agent's actions depend on its reasoning when the correct action depends on a changing state. In a small file-system task,…
Detection Augmented Learning In Non-Stationary Bandit Environments
We study the problem of non-stationary bandits without knowledge of the environmental changes. We propose Detection Augmented Learning (DAL) a black-box paradigm that comprises three modular components: a bandit algorithm, a change detector, and a forced exploration schedule.…
SAR: Learning Global Item-Space Matching and Geometry-Aware Placement Ranking for Buffered Online 3D Bin Packing
AI summaryThe paper introduces Spatial Affordance Retrieval (SAR), a framework for online 3D bin packing that decouples global item-space matching from geometry-aware placement ranking. SAR improves space utilization over strong baselines by up to 4.20 percentage points across standard datasets.
Fair Regression under Approximate Demographic Parity
Exact demographic parity requires the same prediction distribution across sensitive groups. In regression, this can be too restrictive and reduce predictive accuracy. We instead bound every pairwise Wasserstein distance between group-specific prediction distributions by a chosen…
Sector-Ring Linear Reservoirs: Sector-Wise Decay-Timescale Allocation With State Reuse Across Candidate Settings
In reservoir computing, performance depends on whether the fixed dynamics of the reservoir suit the task, and the appropriate memory length and frequency response differ across tasks. The reservoir settings are found by a search on each task, and driving the reservoir with the…
Communication-Efficient Federated Domain Generalization in Edge Networks via Dynamic Prototype Regulation
AI summaryDPGR‑GCAM enables federated Vision‑Language training on edge IoT by combining dual‑parameter‑efficient fine‑tuning with a tiered, channel‑aware transmission that truncates payloads, and a Gradient‑Conflict‑Aware Modulation that mitigates client drift. It improves cross‑domain generalization and converges with only 10 % client participation under fading.
Is it RL or Data? A Controlled Comparison of GRPO and Rejection-Sampling Fine-Tuning
Bauer et al. (2026) showed that in low-data reinforcement learning with verifiable rewards, training GRPO on mixed-difficulty prompts gives up to 5x the sample efficiency of easy-only prompts. They did not test whether this benefit is specific to GRPO, or whether another…
Diffusion Language Models Change Speed Across Tasks and Languages
AI summaryDiffusion language models vary decoding cost across tasks and languages; generating new, rare content requires many forward passes while copying or repeating existing reasoning is cheap. Cost correlates with correctness and can guide efficient inference.
SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation
Training high-quality diffusion models is computationally expensive, and inference-time guidance often adds further sampling cost. In this work, we introduce Swift, a simple yet effective self-guided training method that uses a frozen weak denoiser to improve a stronger model…
Measurement Validity and Representation Dynamics of Confidence Signals in Large Language Models
AI summaryThe paper audits Qwen2.5 models’ confidence signals (verbal, probe-based, sampling agreement) on retrieval and reasoning tasks, finding probes signal agreement earlier for retrieval, verbal confidence unstable across prompts, and sampling agreement offers higher resolution despite higher ECE.
ACE-Trader: Adversarial Co-Evolution for Generator-Critic Trading Systems
Generator--critic trading systems pair an LLM-based trader that makes trading decisions with a red team agent that evaluates these decisions and provides feedback for revision. However, improving individual trading decisions does not ensure that both agents learn from experience…
Rubrics Beyond Evaluation: Exploiting Rubric Structure in Reinforcement Learning
Rubrics are widely used to evaluate open-ended question answering by breaking overall quality into explicit criteria. These explicit criteria make rubrics a rich source of learning guidance, yet their potential for guiding the training process has been largely overlooked.…
Statistical Learning Theory for Direct Preference Optimization
Direct Preference Optimization (DPO) learns from pairwise preferences through a logistic objective whose score-difference parameterization follows the Bradley--Terry model. We study how failure of this modeling assumption limits DPO, distinguishing structural misspecification…
Are Self-Evolving Agents Continual Learners?
Large language model agents are often deployed in environments where new knowledge, tools, and workflows keep arriving. Such an agent is therefore expected to adapt continually, learning new tasks as they arrive without forgetting earlier ones. This resembles the continual…
Adaptive Decoding Methods for Token-Efficient Reinforcement Learning
Rollout decoding affects both which trajectories are used for reinforcement learning with verifiable rewards (RLVR) and how many completion tokens are spent collecting them. We study whether candidate-token support should remain fixed or widen during training in the context of…
CODA: Cohort- and Drift-aware Foundation Model for Multimodal Clinical Reasoning
Recent advances in multimodal foundation models have demonstrated strong performance across diverse clinical tasks. However, existing approaches predominantly operate at the individual sample level and fail to model the interplay between population-level cohort patterns and…
Residual-Guided Adaptive Tucker Rank Allocation for Parameter-Efficient MCI Conversion Prediction
Allocating limited trainable capacity across convolutional layers is a central challenge in parameter-efficient fine-tuning (PEFT) of pretrained three-dimensional (3D) networks. We propose Residual-Guided Adaptive Tucker Rank Allocation, which jointly selects input- and…
Learning to Reason Efficiently with Interventional Step Credit
Large reasoning models often produce lengthy chain-of-thought (CoT) traces mixing useful intermediate steps with unnecessary ones. Recent methods explore step-level signals such as attention mass and step entropy to identify and suppress unnecessary steps. Despite their…
DeltaTrace: Efficient Signed Attribution for Reasoning Language Models
We study how words in the context support or oppose a language model's response and present DeltaTrace, a novel attribution framework that explains the effect of input tokens on a fixed model response. DeltaTrace measures how the response's log-probability changes when the…
Asynchronous On-Policy Self-Distillation
Unlike on-policy distillation (OPD), which depends on a separate teacher model, on-policy self-distillation (OPSD) leverages a single language model acting simultaneously as the student and the teacher, with the teacher conditioned on privileged information. Despite this…
Learn the Interface, Not the Implementation: Output Representation Dominates Training Regime in LLM-Based RTL Generation
Large-language-model register-transfer-level (RTL) generation has concentrated on how to train the model: fine-tuning, continued pre-training, and reinforcement learning from a verifier. We argue this is the second-order question. On a bounded, formally specifiable task of…
Motor compound fault diagnosis using an attention-based cross-talk architecture
As physical AI increasingly relies on sensing to understand physical systems, reliable condition monitoring of industrial machinery has become more important. Motors are ubiquitous in industrial systems, and increasing system complexity raises the likelihood of simultaneous…
Out-of-Distribution Detection in Continual Learning
Out-of-distribution detection in continual learning (OOD-CL) is highly challenging, as it requires (i) learning a sequence of tasks without catastrophic forgetting, and (ii) identifying unknown samples (OOD) without access to known (in-distribution (ID)) data from past tasks,…
An Analysis of Semantic and Spatio-Temporal Co-evolution for Edge Classification on Dynamic Text-Attributed Graphs
Dynamic edge classification on text-attributed Web graphs, such as review networks, community forums, and news-derived knowledge graphs, requires high quality, adaptive semantic representations in the context of network evolution. Existing dynamic graph architectures typically…
Maestro: Agentic Formal Proof Development by Coordination
Lean is increasingly used to verify mathematical proofs and software produced by LLM agents, but writing formal proofs remains hard. Existing prover systems take a divide-and-conquer approach that recursively decomposes hard proof obligations into easier lemmas and splits again…
UtilityBench: Measuring Language Model Utility When Errors Have Different Costs
As language models are increasingly deployed in real-world scenarios with important outcomes, the impact of their mistakes often extends far beyond what correctness can capture. In domains like medicine, a single error can lead to serious consequences and put patient safety at…
EvoSkill v2: Fully Agentic Skill Evolution from Benchmark Adaptation to a Stateful Evolver
EvoSkill v2 is a three-agent system that edits the instructions (skills) an agent is given, leaving its model and harness intact. The user starts a run by describing a benchmark and supplying a grader, which a benchmark adaptor turns into a benchmark-specific plugin. A solver…
Keep the Lineages Alive: Diversity-Preserving SMC for Diffusion LLM Decoding
Inference-time sequential Monte Carlo (SMC) decoding can improve the reasoning accuracy of masked diffusion language models (diffusion LLMs) without post-training by resampling a population of partial generations according to a reward. Its effectiveness therefore depends on the…
The KV-Surgeon: Compensation-Aware Offline KV-Cache Compaction
Offline compaction of the key--value cache (KV-cache) reduces the cost of repeatedly querying a fixed context by constructing a smaller cache once and reusing it subsequently. The central challenge of compaction is deciding which cache entries to delete and which to retain.…
MITRA-lexicon: Lexica of attested translations from Sanskrit into English, Tibetan and Chinese
AI summaryMITRA‑lexicon provides 2.18 million Sanskrit headword entries with attested translations into English, Tibetan, and Chinese, each linked to digitized source sentences and aligned at the span level using a fine‑tuned annotation pipeline. The resource outperforms existing dictionaries on benchmark evaluations.
OrderGPT: Generation Order as a Learnable Modeling Choice
Generative modeling of discrete data has two popular paradigms: autoregressive models such as GPT, which factorize a sequence in a left-to-right (L2R) order, and masked diffusion models, which are essentially any-order autoregressive modeling. Rather than competing methods,…
Productive State-Space Exploration for Multi-turn RL in LLM Agents
Reinforcement Learning (RL) for Large Language Model (LLM) agents in multi-turn, long-horizon environments remains challenging due to inefficient action selection and repetitive behavioral patterns. In particular, agents often become trapped in high-confidence yet unproductive…
Correctness-Dependent Asymmetric Target Distributions for Self-Distillation Policy Optimization
AI summaryAPSD mixes student and self‑teacher distributions based on rollout correctness, preserving good behavior on correct rollouts and sharpening corrections on incorrect ones. It matches or exceeds SDPO performance while reducing training steps by 2.6× on Qwen3 models.
NIOCIS: Typed-Decision Next-Hop Routing with Node-Conditioned Option Spaces for Tool-Use Agents
Tool-use agents pay per decision: at each hop a language model re-reads the state and thinks for seconds, though its actions are finite and decisions recur. We present NIOCIS, a routing layer that folds execution logs into a loop-collapsed action graph (forward transitions plus…
Graph-to-Table Interfaces Control Relational Generalisation
Graph-to-table systems combine three design choices: an encoder that maps relational evidence to feature rows, a context constructor that selects labelled examples, and a tabular decoder. We investigate whether these components can be evaluated independently through matched…
Federated Community Detection in Bipartite Networks
We study federated spectral estimation when a large matrix is horizontally partitioned across servers that cannot share raw data. We propose FLICKER, a target-specific gradient-based power iteration that recovers a singular direction associated with a prescribed spectral target…
Internalizing Chain-of-thought into Looped Transformers
AI summaryThis paper builds constant-bit-size looped transformers that blend chain‑of‑thought and latent‑state computation, achieving a Turing‑machine simulation with context windows and generated tokens each reduced by a factor of L compared to unlooped models.
TempoMoE: Identifying and Steering Fast-Slow Thinking Experts for Efficient Multimodal Reasoning
Multimodal Mixture-of-Experts (MoE) reasoning models rely on long chain-of-thought (CoT) for complex tasks, yet often overthink easy inputs, increasing token cost and latency, and sometimes degrading accuracy. Existing methods optimize reasoning length during training or impose…
PnP-CF: A Consensus Flow-Matching Solver for General Inverse Problems
Plug-and-Play (PnP) methods using continuous-time deep generative models as learned data distribution priors enable synthesizing high-quality solutions of model-based inverse problems. The quality of the solutions depends critically on the interplay and consensus between the…
SPLASH: Contrastive Audit Pressure for Detecting and Localizing Security Hallucinations in Generated Code
Code generated by large language models often passes every functional test while remaining insecure. These security hallucinations are common, written with high confidence, and repeated across samples, so neither the generator's own uncertainty nor static analyzers reliably…
Scenario Generalization and Sector Transfer in Real Surface Irrigation Surrogate Modelling with Tabular Models
Surface irrigation planning requires screening many operating conditions and spatial configurations, while high-fidelity shallow-water-equation (SWE) simulations are expensive. We study tabular surrogate models for predicting maximum water height from drone-derived field…
ProtGlycanDock: Towards Accurate Protein-Glycan Docking with Tailored Dataset, Benchmark and Model
Performing accurate protein-glycan docking is critical to understanding protein-glycan interactions which govern many biological processes. However, this important task is hampered by the lack of tailored datasets, benchmarks and models for protein-glycan docking. To fill this…
Towards instance-dependent regret optimality in Episodic MDPs with Posterior Sampling
We study regret minimization in finite-horizon episodic Markov Decision Processes (MDPs). While minimax-optimal algorithms are known, tractable approaches to instance-dependent optimality are still lacking. Motivated by this gap, we introduce $\pi_0$-PSRL, a variant of PSRL that…
Truncate or Not? A Scaling Study of Looped Transformers with Truncated Backpropagation
Looped transformers reuse a shared block of layers to increase computational depth without increasing the number of trainable parameters. They are trained with backpropagation through time (BPTT). To save training compute, BPTT is often truncated to the last few loops, although…
FinHopper: A Progressive Multi-Hop Benchmark for Complex Financial Agents with Step-wise Evaluation
Financial agents are expected to answer factual investment queries through open search, but financial multi-hop retrieval differs from general multi-hop QA in a crucial way: each intermediate answer is an exact typed constraint, such as an entity, date, event, or numerical…
LeakBench: Auditing Reward Hacking through Prediction-Time Leakage in Clinical Modeling Agents
Autonomous clinical-modeling agents can optimize predictive performance while violating conditions that make a model usable. We introduce LeakBench, an executable, guideline-derived audit suite that tests agent decisions for prediction-time leakage, calibration, site-level…
Multilingual Voicemail and Answering Machine Detection via Pretrained Audio Embeddings and Gradient Boosting Machines
AI summaryLightweight voicemail detection combines pretrained audio embeddings with gradient‑boosting, achieving top precision and low latency across multilingual data. It also attains 94.34% robocall recall on unseen data and outperforms proprietary, academic, and LLM baselines.
Provable Adversarial Detection: Prime Quantization meets Gromov--Wasserstein
Adversarial vulnerability persists across modern vision architectures from CNNs to vision language models (VLMs), yet existing detection methods rely on heuristics without theoretical guarantees. We address the fundamental question of when adversarial perturbations can be…
Learning from a Bandit Learner
Classical inverse bandit methods typically aim to recover true reward parameters. In contrast, we infer a demonstrator's subjective beliefs, specifically its internal estimates of suboptimality gaps. We consider no-regret demonstrators that follow randomized policies, such as…
Declarative Explanations for Graph Neural Networks: A Demonstration
Graph Neural Networks (GNNs) have proven effective for graph analytical tasks, yet their "black-box" nature complicates the understanding of their layer-wise behavior and hinders their application to large graphs. We demonstrate SliceGXQ, an end-to-end declarative and…
From Retrieval to Recognition:Knowledge-Enhanced Post-training of Time Series Foundation Models for Classification
Time series foundation models (TSFMs) have achieved notable progress in classification tasks through large-scale pretraining. However, their post-training practice mainly focuses on implicit parameter fine-tuning, where task-relevant information is entangled with abundant…
Variance-Aware Non-Stationary Bandits:All-Budget Regret Bounds and Adaptation Costs
Low reward variance reduces estimation cost, but does not eliminate monitoring of arms that may improve while unobserved. We refine the variance--variation tradeoff and characterize how both costs depend on supplied variance and variation budgets in bounded, non-stationary…
Optimal Multi-Kernel Combination for Online Multi-Label Classification
Multi-label classification is particularly challenging in streaming scenarios, where data arrives sequentially. Online kernel methods offer a promising solution. However, prevailing methods rely on a single preselected kernel, making them prone to suboptimal choices since the…
Nash-MOPD: Optimizing Task Trade-offs in Multi-Teacher On-Policy Distillation
AI summaryNash-MOPD schedules teacher sampling by solving a local minimax game over reverse‑KL statistics, giving more prompts to domains that improve slowly. It outperforms fixed and D³‑MOPD, achieving higher math and code scores on DeepSeek‑R1‑Distill‑Qwen‑1.5B and Qwen3‑8B‑Base.
Reinforcement Learning Controllers for Fusion Scenario Development
Scenario development, the design of the waveforms that take a tokamak discharge from breakdown to flat-top, is still an iterative, labour-intensive process. Every candidate scenario has to be simulated with a controller in the loop, and the classical shape and current…
FLMNet: Self-Ensemble Distillation from Masks to Seeds for Single-Tower CTR Prediction
Click-through rate (CTR) prediction underpins recommender systems and online advertising, where high-dimensional categorical features interact in combinatorially many ways. Several recent CTR models, such as FinalNet and QNN, report higher accuracy when a network also learns…
Physics-Driven Framework for QoS-Constrained Satellite System with Non-Linear PA Distortion
This work develops a physics-driven self-supervised graph learning framework that directly optimizes the distortion-aware precoder under nonlinear correlated power amplifier (PA) distortion, without relying on expensive optimizer-generated labels for a user-centric satellite…
Dynamical Batching: How Class-Cycling Training Shapes the Internal Structure of Neural Networks
We investigate whether cyclically varying the class composition of training batches can modify both the performance and the internal organization of neural networks. Motivated by nonequilibrium systems, in which cycling between related sub-goals can make switching between their…
TaQFormer: A Treatment-as-Query Transformer with Data-Fusion for Multidimensional Treatment Modeling in Coupon Allocation
We study coupon allocation on an online marketplace offering multiple service categories. Issuing coupons requires choosing a vector of category-specific discounts, accounting for both their direct effects and cross-category demand shifts. We present TaQFormer, a…
OceanTensorCube and ST-MMT: A Multimodal Data and Transformer System for Seven-Day Forecasting of Pyropia Bleaching
Bleaching of cultured Pyropia (laver) causes losses in Korean seaweed farming. We present a system that forecasts bleaching seven days ahead for farm clusters. OceanTensorCube aligns 32 channels from public marine, meteorological, ocean-model and satellite sources on a 30 m…
RECAP: Reciprocal Sparse Attention and Hierarchical Caching for Efficient Video Diffusion
AI summaryRECAP couples sparse attention with hierarchical caching, using reciprocal correction guided by online error feedback to share refinement information across denoising steps, achieving 3.05× speedup on HunyuanVideo 720p 5 s while improving fidelity over existing sparse‑attention and caching methods.
SEAM: Spectral Embedding Alignment and Merging for Low-Resource TTS
AI summarySEAM expands TTS vocabularies without audio by spectrally filtering subtoken embeddings and aligning them with a pretrained manifold via ridge regression. It matches or exceeds baselines using up to 4× less data, cutting sample complexity by ~75 %.
A Pitfall in the Evaluation of Gene Perturbation Models
AI summaryThe paper shows that common benchmarking practices for gene perturbation models inflate significance estimates by replicating a single prediction across cells, causing unstable precision rankings that vary 86.7% of the time; correcting this reveals a random‑forest model with Gene Ontology features as best overall.
Trust Your Targets: Is Target Propagation Block-Diagonal Gauss-Newton?
Target propagation (TP) is a backpropagation alternative in which local targets carry the global learning signal. Prior works claim that TP performs Gauss-Newton (GN) optimisation on these targets, but the correspondence has only been established under unrealistic assumptions…
Learning SOCP Solver Dynamics with Graph Neural Networks
Second-order cone programs (SOCPs) are a fundamental class of convex optimization problems widely applied across engineering domains. Classical interior-point methods are robust and accurate, but their iterative nature can be costly in real-time or repeated-solve settings. This…
Toward Online Robust Zero-Sum Markov Games with Function Approximation
Online robust zero-sum Markov games pose a coupled operator-learning and game-solving problem: from nominal interaction data, the learner must recover a worst-case minimax Bellman operator while maintaining strategic coverage and controlling stage-game approximation error.…
CAST: Cast-Admissible Generative Design and Shape Optimization with Paired Depth Images
Deep generative design can produce complex geometries, but unconstrained representations may create undercuts that prevent release from a two-part mold. We present CAST, which reformulates three-dimensional shape generation under a selected mold-opening constraint as a…
TriScore: Post-Hoc Out-of-Distribution Detection with Energy, Boundary Probes, and Transform Consistency
Post-hoc out-of-distribution detection usually relies on a single signal extracted from a frozen classifier, such as confidence, energy, features, or gradients. This dependence becomes fragile under classifier-level adversarial shift, where the shift can preserve the detector's…
The Control Layer Behind Rule-Following Agents Trading Across Market Conditions
Trading discipline often fails people. Our study tested whether four AI trading agents followed their strategies and risk rules. Offering traders the evidence and an agent evaluation toolkit to separate luck from following the rules, before giving agents meaningful autonomy or…
Typed Decision Models under Prevalence Shift: Decision Training, Calibration, and Implicit Priors
Typed decision models (TDMs) return a probability over caller-defined options, and Jev is described by its vendor as calibrated. But in deployment class frequencies change, so calibration measured on a fixed test set need not hold where thresholds are applied. Existing…
CIS-Transolver: A Cross-Layer Incremental-State Neural Operator for Fine-Scale Automotive Aerodynamic Flow Prediction
Neural operators offer an efficient data-driven approach to rapid prediction of complex automotive aerodynamic flows. However, spectral bias and other factors limit accurate reconstruction of fine-scale flow variations near complex geometries. Existing studies improve the…
Logit Averaging During Training Sets a New State of the Art for Non-Invasive Speech BCI
Non-invasive brain-computer interfaces are in practice limited by signal-to-noise ratio, and recent state-of-the-art methods address this limitation by aggregating over repeated trials of the same stimulus. Yet multi-trial aggregation has been treated inconsistently across the…
PhyPro-FM: A Physics and Prompt-Guided Graph Foundation Model for Contingency Analysis
AI summaryPhyPro‑FM combines graph neural networks with a learnable global query and prompt‑guided few‑shot learning to perform contingency analysis. It outperforms prior methods by prioritizing branch‑flow errors during training and using a residual‑based correction for singular branches, achieving superior accuracy and robustness.
An Empirical Study of Parameter-Efficient Fine-Tuning for Mixture-of-Experts LLMs
Parameter-efficient fine-tuning (PEFT) methods are critical for adapting large models under resource-constrained settings. However, as mixture-of-experts (MoE) architectures, which offer efficient inference at massive parameter scales, gain increasing prevalence in the…
CTLite: Lightweight Irregular Multivariate Time Series Forecasting with Reusable Continuous-Time Summaries
Irregular multivariate time series forecasting (IMTSF) requires capturing temporal dynamics and cross-variable dependencies from sparse, asynchronous observations. Existing methods often rely on learned temporal encoders, coupling historical representation construction with…
Selecting the Best Few with Human Evaluators: Adaptively Combining Ratings and Pairwise Comparisons
This paper considers a decision-maker tasked with identifying the $k$ best items from a large pool of candidates, such as job applicants, LLMs, award nominees, or new product ideas. The decision-maker learns about the items' quality by soliciting human evaluations. Each…
ForagerSim: Evaluating Embodied Map-Based Reasoning in Large-Scale Dynamic Worlds
AI summaryForagerSim is a modular simulation platform that tests embodied agents on map-based reasoning tasks across diverse scales, dynamics, and perception settings. The study shows that while simple 2D sensing lets learned policies succeed in small environments, humans and map-based baselines outperform them in larger, complex topologies.
LaMER: Language-Conditioned Multi-Expert Routing for Weakly Supervised Referring Expression Comprehension
Referring Expression Comprehension (REC) localizes the image region described by a natural language expression, underpinning applications such as autonomous navigation, human-robot interaction, and medical image analysis. Under weak supervision, REC is challenged by the…
A Simple and General Approach to top-$k$ Selection
This paper considers the Bayesian top-$k$ identification problem, wherein a decision maker is tasked with selecting $k$ items from a pool of items of unknown quality. To assess the items, the decision maker adaptively allocates a fixed sampling budget, where each sample of an…
Co-Evolving Policy and Reward for Code Generation: A Mechanistic Study of Single-Adapter Drift
Reinforcement learning with verifiable rewards (RLVR) is particularly well suited to code generation because generated programs can be objectively evaluated through execution. Execution also provides behavioral information that can be reused as additional supervision. The…
Local data augmentation for imbalanced binary classification
Facing imbalanced binary classification, we aim at maximizing the recall while limiting the proportion of false detections in the minority class. For doing that, the present work studies the effect of several rebalancing strategies, from the uniform subsampling of the…
R2-GP: Variance-Calibrated Priors for Additive Gaussian Processes
We develop R2-GP, an R2D2-type prior for Gaussian process (GP) generalised additive models (GAMs) that calibrates explained variance ($R^2$) for Matérn covariance kernels. The function variance for standard Matérn kernels depends jointly on the effect magnitude and length-scale.…
Rethinking LLM Fine-Tuning via Weight Space Reparameterization: Preserving Safety during Downstream Adaptation
Fine-tuning large language models (LLMs) on downstream tasks often weakens their previously aligned safety behavior, creating an inherent trade-off between task adaptation and safety preservation. Existing approaches attempt to mitigate this issue by identifying safety-relevant…
Regime-Aware Interpretable Ranking for Sector Rotation
Sector rotation is a common decision in benchmark-relative portfolio management, where portfolio managers rank sectors for overweight and underweight positions. We formulate this task as ranking sector actions by expected future risk-adjusted active performance, measured with…
Residual-Based Adaptive Refinement in Hybrid Variational-Collocation PINNs for Singular Elliptic PDEs with Mixed Dirichlet-Neumann Boundary Conditions
Solutions of second-order elliptic PDE problems on nonconvex or cracked domains with mixed Dirichlet-Neumann conditions typically lie outside $H^2(\Omega)$. Near re-entrant corners, crack tips and points where the boundary condition changes type, the gradient grows like…
VeriCUA: Learning Proactive Verification for Reliable GUI Agent Rewards
Reinforcement learning for computer-use agents increasingly relies on proactive reward agents that interact with the post-execution GUI environment to verify what recorded screenshots cannot reveal. The reliability of such verification hinges on two coupled decisions: what to…
On Learning Expand-and-Sparsify Representations
Expand-and-sparsify ($\texttt{EaS}$) representations lift an input to a much higher dimension through a linear map and keep only the $k$ largest coordinates. Existing theory builds that map at random and needs the expanded dimension to grow exponentially in the input dimension,…
A Simple and Unified Method for Gradient-Variation Regret
Gradient-variation online learning has attracted increasing attention for its connections to game theory and optimization. It has been studied in both non-stationary online learning, which aims to minimize dynamic regret against time-varying comparator sequences, and universal…
Wet-Biased Data Pipelines Mislead the Training and Evaluation of Generative Precipitation Nowcasting
Generative models for precipitation nowcasting are trained and evaluated predominantly on wet-biased datasets, which over-represent precipitation events to emphasize rare extremes. Selecting test cases based on target observations is known to favor forecasts that match the…
Strength Controls Change How Interventions Compare Across Sparse Autoencoder Codes
AI summaryThis paper evaluates how different strength controls, such as norm matching and output calibration, affect sparse autoencoder intervention comparisons in language models. It shows that full-output calibration preserves transfer decisions better than total-norm matching.
Learn the Task, Not the Schedule: Reducing Spurious Cross-Arm Coupling in Bimanual Imitation Learning
A bimanual demonstration specifies both what the robot should do and one particular schedule for when each arm should act. Standard imitation learning does not distinguish task-required cross-arm constraints from incidental scheduling choices, and may therefore exploit…
RouterWorld: Scaling Conversational AI with Latency-Aware LLM Routing
AI summaryRouterWorld benchmarks conversational LLM routing with realistic latency and partial observability, offering 371k responses and 76k agent trajectories. It shows most existing routers prioritize accuracy over latency when only sparse logs or streaming data are available.
A Fisher-Geometric View of Multi-Teacher On-Policy Distillation
Consolidating specialized foundation models into a unified generalist is a central goal of post-training. Two approaches are in common use: model merging, which is fast but treated as a heuristic, and multi-teacher on-policy distillation, which matches distributions but demands…
A Semantic Knowledge Graph Construction Pipeline for Vehicle Intention Prediction in nuScenes
Knowledge graphs (KGs) are increasingly used in autonomous driving (AD) to make traffic scene entities and relations explicit for prediction and reasoning. Existing traffic-scene KGs, however, primarily emphasize trajectory forecasting or general scene understanding. This paper…
Training Natural Schemers: RL Can Amplify Oversight-Conditional Behavior
One of the central concerns for the safety of advanced AI is scheming: when a model strategically complies under oversight but pursues its own goals when unobserved. Current demonstrations, such as Anthropic's Alignment Faking work, depend on design choices that make them less…
SafeWall: A Risk-Aware Job Walltime Recommendation for HPC Scheduling
In high-performance computing (HPC), users must submit resource requirements for their jobs, with user-supplied walltime being a critical parameter. However, these estimates are notoriously inaccurate, as users tend to overestimate walltime to prevent the scheduler from…
WhatIf-Bench: Benchmarking Intervention-Aware Time Series Forecasting
Recent time series foundation models (TSFMs) can consume covariates in-context, extending zero-shot forecasting to problems driven by external factors. This enables scenario analysis: forecasting how a target responds to events that set covariates to chosen future values.…
Concerns in Benchmarking Jailbreaks: An Empirical Study of Attack Success Rates
Jailbreak methods are commonly compared using Attack Success Rate (ASR), but these comparisons depend on how responses are generated and judged. In this paper, we examine eight attacks against three open-weight language models using three evaluation procedures. Low-resource…
Bombadil: A Foundation Model for Plant Electrophysiology
Living plants, like animals, use electrical signals to carry information about their physiological state across their body, from roots to branches. These signals can be recorded, as they are in animals, from a pair of differential extracellular electrodes inserted into…
ITU-DESC: a missingness-aware machine-learning dementia risk score in the UK Biobank cohort
Background: Existing dementia risk scores rely on fixed, pre-selected features and linear models, or ignore bias from informative missingness in cohort data. Here we introduce the Istanbul Technical University Dementia Risk Score (ITU-DESC), an explainable machine-learning score…
A Probabilistic, Rotation-Preserving Kernel Operator Surrogate for Continuum-Robot Design Spaces
We develop an uncertainty-aware, operator-learning-based surrogate for tendon-driven continuum robots (TDCRs) that generalizes across robot design and actuation while predicting the complete equilibrium configuration. Our central idea is a nonstationary Gaussian-process (GP)…
Degree Bounds for Separating Polynomial and Neural Invariants for Point Clouds
Neural networks that respect the symmetries of point clouds commonly compute high-order representations of the rotation group to improve their expressivity. Such models are known to be universal, and thus able to separate any pair of point clouds, when using representations of…
Global Time-Aware Conditioning (GTC): A Mechanism for Time-Aware Pediatric Sleep Representation Learning
Sleep physiology evolves throughout the night, with changes in sleep cycles, respiratory stability, and cardiovascular dynamics. However, self-supervised learning (SSL) methods for sleep typically process short, fixed-duration windows without explicitly modeling where each…
SleepPairNet: Learning Multimodal Pediatric Sleep Representations with Random-Pair Cross-Attention
Analyzing physiological sleep signals is important for assessing sleep quality and detecting sleep-related disorders. Learning inter-modal relationships across physiological signals can provide richer representations for downstream sleep analysis. We introduce…
VaDar: Implicit Goal Conditioning via Value Disaggregation
Long-horizon tasks are important for modern AI and robotics, yet pose a challenge for reinforcement learning. Historically, a popular approach has been to decompose a task into subgoals, and learn a subgoal-conditioned policy to execute each subtask in sequence. We introduce an…
When the harness decides success: state that cannot be re-observed
Agent harnesses decide what a language model sees across turns: they drop old turns, summarize them, or replay the model's own earlier reasoning. We ask when these choices change whether an agent succeeds, not only what it costs, and report three findings. First, in a…
The Effectiveness and Durability of Pretraining Data Filtering in Protein Language Models
Pretraining filtering is one of the most common safeguards for open-weight language models. Here we study its effectiveness, durability and scalability using the ESM-2 family of protein language models. We find that filtering viral proteins from the pretraining data meaningfully…
The Development of Sycophancy in Language Models: Tracing Vulnerabilities from Pretraining to Post-Training
Language models can change a committed answer to accommodate a user's opinion or pushback without new evidence. However, when this behaviour develops during training and what a released model inherits from earlier stages remain unclear. We follow 741 checkpoints of 30 open model…
Nested Shared-Operator Recursion on ARC-AGI-1: An Empirical Study of HRM, TRM and URM
Recursive reasoning models refine persistent internal states by repeatedly applying learned operators. We investigate whether adding a persistent intermediate state that integrates repeated updates of a faster state through the same operator improves predictive performance…
PACT: Plan-Augmented Coordination via Trace-Graphs for Dynamic Multi-Agent Combinatorial Optimization
AI summaryThis paper introduces PACT, a multi-agent reinforcement learning method for dynamic combinatorial optimization that uses trace-graphs to coordinate agents as problem information is progressively revealed. PACT improves solution quality over baselines and successfully scales to larger instances, even outperforming centralized oracles with privileged information.
Which Model for Which Role? Cost-Efficient Pareto Recommendation for LLM Agentic Workflows
LLM agentic workflows increasingly employ different models for specialized roles. Users may prefer different trade-offs between task quality and deployment cost, sometimes also considering latency. Finding good assignments incurs a separate search cost from evaluating multiple…
Intra-Rubric Reward Divergence: Mitigating Rubric Hacking through Latent Dimension Recovery and Balancing
Rubric hacking limits both performance and training robustness in rubric-based reinforcement learning. By decomposing rubrics into sub-dimensions, we reveal the Intra-Rubric Reward Divergence: as the total rubric score increases, certain sub-dimension scores systematically…
Parallel Language Models via Variational Decoding
AI summaryParallel Language Models generate multiple tokens per forward pass via latent auxiliary variables, trained with a variational objective that combines autoregressive and parallel decoding losses. On 80 M–3 B models, PLMs match autoregressive quality while achieving 2.3×–5.7× inference speedups.
GeoSplit: Selective Edge--Server Map Acquisition for Cross-User Travel-Mode Recognition
Networked mobility sensing couples local trajectory inference with remote geospatial context, making communication decisions part of the prediction system. We present GeoSplit, an edge--server framework for travel-mode recognition that separates the utility of map evidence, its…
Doubly Robust Learning of Heterogeneous Survival Treatment Effects under Time-Varying Treatments
Estimating heterogeneous survival treatment effects (HSTEs) from time-to-event data under time-varying treatments is important in fields such as medicine (e.g., to decide which sequence of treatments is better in terms of survival probability). However, this task is challenging…
Deep Pontryagin Learning for Continuous-Time Stochastic Control with Jumps
We develop a deep Pontryagin learning framework for high-dimensional continuous-time stochastic control problems that naturally accommodates jump-diffusions. In contrast to direct policy optimization, which repeatedly differentiates the objective through the entire…
Copying Under Repetition: From Prefix Matching to Position Hashing
Frontier LLMs can copy random text almost flawlessly, yet they often overrun or truncate long blocks of repeated characters. In repeated strings, matching local content no longer reliably identifies where copying should continue or when it should stop. Nevertheless, we show that…
Injection of Space Radiation Errors on Low-power NVM using Generative Deep Learning
We develop an error model for a Non-Volatile Memory (NVM) device that represents the effects of physical faults experienced in a CubeSat mission in Low Earth Orbit (LEO). The CubeSat carries an off-the-shelf low-power Ferroelectric RAM (FRAM) chip and runs repeated read/write…
Cross-Fitting for Neural Posterior Estimation
Neural Posterior Estimation (NPE) fits an amortized conditional density estimator to a simulated training set of complete-data realizations. NPE's training objective, log-likelihood, favors posterior approximations that assign arbitrarily high density to the training points, and…
CAVA: Complementary-Label-Aware Data Valuation
Complementary-label learning (CLL) trains a classifier from labels that name a class an instance does not belong to. Such labels are cheap to collect from crowdworkers or vision-language models (VLMs), making them attractive for large-scale annotation. However, both sources can…
Polar Switch: Fast Muon Orthogonalization with Polynomial Circuits
AI summaryPolar Switch introduces low‑FLOP polynomial circuits for orthogonalizing gradient matrices, achieving 15–18% faster runtime than Polar Express and 8–16% faster than Gram Newton–Schulz while maintaining comparable accuracy and numerical stability in large‑scale language‑model training.
recAMPLIFY: Large-Scale Pretraining of Recursive Protein Language Models
AI summaryrecAMPLIFY trains a recursive masked protein language model with 155 M unique parameters on 2 T tokens, achieving a 22.5% lift in fold accuracy and better contact recovery while using 56% fewer parameters. Gated Mixture Recurrence (GMR) adds adaptive gating, reducing expected depth by up to 37% with modest accuracy loss.
Gradient descent reaches arbitrary accuracy in a fixed number of large steps on multiclass logistic regression
AI summaryGradient descent with a suitably large step size attains any desired accuracy on multiclass logistic regression in a constant number of iterations independent of ε. The required step size depends on ε and dataset properties, while the iteration bound depends only on data size and margin.
Trajectory Replay: Continuous-Time Mixture Transport for Lifelong Generative Models
Adapting pre-trained continuous-time generative models to sequentially incoming data without forgetting earlier tasks is a central challenge in lifelong learning. A leading strategy, generative replay, uses a generative model trained on historical tasks to augment its own…
Simulation Priors for Rapid Adaptation in Model-Based RL for Autonomous Racing
Reinforcement learning enables learning high-performance control policies, but its reliance on extensive interaction limits deployment on physical systems where data collection is costly, and failures are unsafe. We introduce Prior-Informed Adaptation (PIA), an offline-to-online…
HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs
Harness development plays a crucial role in building capable AI agents. Recent approaches improve harnesses through recursive self-improvement (RSI), but their performance gains often come with higher execution costs. We introduce HarnessPro, a label-free method for evolving…
Topological Sheaves of Irreducibility: A Theoretical Framework for Multi-Agent Task Obstructions
Multi-agent systems can differ under identical agents and communication budgets when they support different forms of joint interaction. We introduce Topological Sheaves of Irreducibility (TSI), a sheaf-theoretic framework separating task-level compatibility obstructions from…
Cooperative Multi-Agent Reinforcement Learning for Market Making on Prediction-Market Order Books
A market maker provides liquidity by quoting prices to buy and sell an asset, earning the spread while managing inventory risk. We study market makers trained with multi-agent reinforcement learning on prediction markets, specifically Polymarket, whose thin order books pose…
Training with ML-labeled Data via Stochastic First- and Zeroth-Order Optimization
Training a machine learning (ML) model is an optimization process that typically requires a set of examples with ground-truth labels beforehand. Obtaining these ground truths usually requires laborious human effort and can be costly. The scalability of human labeling becomes a…
On Data Scaling in Tabular ML: More Data, Better Generalization?
Recent work in tabular machine learning has emphasized the importance of diverse and realistic evaluation settings. The niche of large-scale (1M+) data, however, remains under-explored. To study training-data scaling, we assemble eight real-world tabular datasets with up to…
Two hidden layer suffice: the caterpillar net
AI summaryCaterpillar Net shows that a ReLU network with two hidden layers can compute the maximum of \(N\) inputs, achieving widths \(O(N^2)\) and \(2^{O(N\log N)}\). This confirms that two layers suffice for any piecewise‑linear maximum, refuting the earlier conjecture of logarithmic depth.
Teacher Supervision, Fixed Representations and Calibration: Exact Risk Calculations for Candidate-Scoring Decision Models
Candidate-scoring decision models return a probability distribution over a presented set of answers. Public documentation of one such system, Jev, and of an open counterpart, JevAny, describes three training ingredients that determine what these probabilities estimate: soft…
Learning to Refer: Inference-Time Collaboration in Decentralized Federated Learning
AI summaryEdgeDRM enables inference-time collaboration in decentralized federated learning by learning local density potentials that guide query referrals along a graph, aggregating predictions from visited clients. It outperforms graph‑matched baselines, especially in sparsely connected settings.
Scaling Parameter-Efficient Fine-Tuning for Brain Encoding
AI summaryPEFT with a low‑rank LoRA adapter lets us adapt 10⁴⁰‑parameter multimodal backbones for brain encoding, improving performance on nine datasets and winning the Algonauts 2025 Challenge. A multi‑LoRA kernel enables an 200‑model ensemble that tops the post‑challenge leaderboard.
Next Best Action: Causal Constrained Offline RL for Business Action Targeting
Next Best Action (NBA) systems must select among customer interventions (e.g., discounts, promotions, and lifecycle actions), maximize long-term economic value, satisfy complex operational constraints, and remain causally grounded. Existing approaches lack a rigorous framework…
MemT-1.0: Towards Memory-Native Evolution
AI summaryMemT-1.0 integrates memory directly into the model, using Gated DeltaNet-2 and Mixture-of-Storage, and is trained on 34B tokens. It outperforms parametric memory baselines, achieving 1.86× faster inference and 66× less storage at 128K token history.
Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
AI summaryWildAWM synthesizes 341 long‑horizon computer‑use tasks in controllable, multi‑interface environments using open‑source models, and training Qwen agents in these worlds boosts performance on eight out‑of‑distribution benchmarks.
ProteinGeoBench: Frontier Structural-Biology Agents Are Fluent at Lookup but Lost in Geometry
AI summaryProteinGeoBench shows frontier models excel at metadata lookup (0.785–0.816) but struggle with numeric geometry (often failing 12/28 tasks); performance varies with evaluation budget, harness, and network access, underscoring the need for per‑task reporting.
Supervised Learning with Quantum Hamiltonian Embeddings
AI summarySupervised learning framework that encodes a dataset into a parameterized Hamiltonian; the ground state defines the classifier. Efficient gradient estimation, generalization bounds, and cryptographic hardness proofs show advantages over classical methods; demonstrated on IBM Heron.
Kronecker Delta Rule for Long Context Linear Attention
AI summaryKronLSTM introduces the Kronecker Delta Rule, a minimally invasive memory‑revision update for linear attention. It converges to a stable steady‑state, keeps the memory norm bounded, and outperforms existing linear attention models on long‑context language tasks.
Newton-Based Methods for Bilevel Optimization Beyond Lower-Level Strong Convexity
AI summaryNewton-based GALET and BAMM solve bilevel problems without requiring lower‑level strong convexity, using shared curvature between lower‑level and multiplier updates; GALET handles singular Hessians under a spectral‑gap condition, BAMM uses a convex aggregation model. Both achieve provable convergence and better complexity than prior methods.
WeAct: Dynamic Weight Switching as Agent Actions
Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories. Updating shared parameters to learn new capabilities can degrade previously acquired ones, resulting in catastrophic forgetting. As a trajectory unfolds, different stages may require…
CoopUQ: Robust Multi-Robot Social Navigation via Uncertainty-Aware Cooperative Forecasting
AI summaryCoopUQ fuses cross‑robot spatiotemporal features to produce uncertainty‑aware occupancy forecasts, then uses these forecasts in a constrained multi‑agent reinforcement learning policy that penalizes traversal of high‑risk or uncertain cells. It shows better prediction accuracy and safer navigation in both in‑distribution and out‑of‑distribution Unity3D and real‑robot tests.
OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory
AI summaryOpenVLHarness adds a tool‑capability layer and persistent multimodal memory, enabling large language models to fetch, interpret, and store heterogeneous evidence across reasoning steps. It boosts performance on 22 benchmarks by 9.7–12.6 points over no‑tool baselines while reducing token cost.
Text-to-Image Generative Modeling Without Language Pretraining
AI summaryHalcyon trains multimodal autoregressive models from scratch on 100M image‑text pairs, surpassing state‑of‑the‑art text‑to‑image systems while using far less pretraining compute and showing language pretraining is unnecessary.