AI News

Why Does Adaptive Batching Help LLM Pretraining? A Perspective from Unbounded Variance (arxiv.org)

Open-Endedness Bench: Measuring Epistemic Process from Agent Records (arxiv.org)

TasteBench: Multimodal Benchmark for Sensory Prediction, from Molecules to Sustainable Foods (arxiv.org)

Time Series Forecasting Benchmarks Need Scenario-Grounded Stress Testing (arxiv.org)

VERSE: Verified Self-Evolving Optimizer for Agent Harnesses (arxiv.org)

JARQ: Joint Alternating Refinement for Quantization (arxiv.org)

Lexicographic Multi-Objective On-Policy Distillation (arxiv.org)

ArrivalBench: Agent-Generated Data Pipelines Are Correct Once and Wrong Under Time (arxiv.org)

Co-design Gym: A Unified Benchmark for Embodiment-Policy Co-optimization (arxiv.org)

Latent-MOPD: Latent Multi-Teacher On-Policy Distillation (arxiv.org)

CuBEs: Culturally-Situated Behavioral Evaluations and the Limitations of Culture-Blind LLM Judges (arxiv.org)

Lost in the Request: How Communication Variation Disrupts Retrieval and Action in Email Agents (arxiv.org)

A GHOST in Long-Horizon Agents: Governance Hazard from Overlooked Safety Constraints across Turns (arxiv.org)

Spend Teacher Tokens Where They Matter: Success-Referenced On-Policy Distillation (arxiv.org)

DataWeave: Deploying Human-LLM Analytics for Exploratory Structured Data Analysis (arxiv.org)

Large language models exhibit unreliable updating of clinical judgment as patient evidence evolves (arxiv.org)

Decoupling Memory from Context: Structured Memory for Token-Efficient Test-Time Continual Learning (arxiv.org)

Learning to Revise Reasoning with Segment-wise On-Policy Distillation (arxiv.org)

Dynamic Expert Pruning for Multi-Agent Systems (arxiv.org)

SlimKV: Joint Token-Feature KV Cache Compression with Reconstruction-Free Beacon Attention (arxiv.org)

Hyperparameter selection for equation learning with biologically-informed neural networks (arxiv.org)

Understanding Trajectory Heterogeneity in Federated World Model Learning (arxiv.org)

Dirac-Interconnected Neural Elements: Discovering Modularity in Physical Systems Without Reduction (arxiv.org)

Differentiable Koopman Operator for Contrastive Learning on Dynamic Graphs (arxiv.org)

Sentry: Learning to Recover from LLM Agent Failures at Test Time (arxiv.org)

Temporal Geometry of Deep Networks: Hyperbolic Representations of Training Dynamics for Intrinsic Explainability (arxiv.org)

Learning Jazz Pianist Style with Cross-Attention Conditioning (arxiv.org)

Label-Efficient Time Series Classification at Scale: A Dual-Stream OSSE-LSTM with Counterfactual Attribution (arxiv.org)

Ego2World: Compiling Egocentric Cooking Videos into Executable Worlds for Belief-State Planning (arxiv.org)

On the Chain-of-Thought Monitorability of Looped Language Models (arxiv.org)

Dynamic LLM Routers are Often Misguided (arxiv.org)

Law And Order: Tax Law Autoformalization (arxiv.org)

Equivariant Visual-Tactile Diffusion Policy for Contact-Rich Manipulation (arxiv.org)

Hybrid Machine Learning-Assisted Raman Spectroscopy with Generative Feature Augmentation for Pharmaceutical Identification (arxiv.org)

The Price of Greenwashing: Algorithmic Verification and Market Discipline using Conformal Machine Learning (arxiv.org)

State-Space Unlearning for Non-Stationary Bias in Land Surface Forecasting (arxiv.org)

Hesitation Has a Geometry: Entropy-Trained Hyperbolic Probes for Sparse Activation Steering (arxiv.org)

Inherit-MAS: Test-Time Evolution of Multi-Agent Systems through Workflow and Execution Inheritance (arxiv.org)

Quantifying the Value of Constructive Induction, Knowledge, and Noise Filtering on Inductive Learning (arxiv.org)

Learning from the Gap Between Pass@K and Pass@1 (arxiv.org)

Copula Active Subspaces I: A Score-Covariance Method for Reduced-Order Non-Gaussian Density Estimation (arxiv.org)

Interpreting at Write Time: A Policy Ablation for Multi-Goal Agent Memory (arxiv.org)

Pragmatic DML with AI-Learned Representations (arxiv.org)

PAPER2LLM++: Continual Self-Evolution of LLMs from Research Papers (arxiv.org)

Modeling Shared and Individual Structure for Cross-Subject Continuous Affect Regression from EEG-fNIRS (arxiv.org)

BitNest: Bit-Nested Speculative Decoding for Memory-Efficient LLM Inference Acceleration (arxiv.org)

VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning (arxiv.org)

ROUTEAUDIT: Interaction-Aware Identification for Budgeted Multi-Verifier Routing (arxiv.org)

iS-KV: Online Low-Rank KV Cache Compression via Block-Incremental SVD (arxiv.org)

MetaRubric: Learning to Reward for Rubric-Based Reinforcement Learning (arxiv.org)