AI News

Progress Reward Model for Reinforcement Learning via Large Language Models (arxiv.org)

GraphACM: Graph-based Action Consistency Measurement for Uncertainty Quantification in LLM Agents (arxiv.org)

Q($\gamma_H$): Entropy-Guided Horizon Mixing in Multi-Discount Q-Learning (arxiv.org)

CARA: Causally Guided Retrieval Agent for Question Answering (arxiv.org)

OntoAct: Hypothesis-Based Neuro-Symbolic Agents through Ontological Abstraction (arxiv.org)

CoWorld Reasoner: Agentic Multimodal Reasoning with the Evolving Video World Model-Driven Harness (arxiv.org)

Jev-LiteMAS: Separating Decisions from Generation in Multi-Agent Workflows (arxiv.org)

Efficient Pairwise Difference Learning: A Comparative Study of Encoding Strategies and Training Pair Selection (arxiv.org)

ReFlow: Replay-Driven Decoupling for Efficient RL Post-Training (arxiv.org)

Local Temporal Contrasts on Human Motion Data Induce a Latent Geometry (arxiv.org)

Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation (arxiv.org)

Predicting Live Birth After Fresh Embryo Transfer in IVF/ICSI: Differential ICM/Trophectoderm Morphology Encoding and a Hybrid Attention--Logistic Regression Ensemble (arxiv.org)

Noise-Space Reward Alignment for Few-Step Generative Models (arxiv.org)

Global Coherence: When Every Agent Is Right and the Team Is Still Wrong - A Local-to-Global Semantic Foundation for Multi-Agent Collaboration (arxiv.org)

Watts If? A Benchmark for Counterfactual Reasoning About the AI Data Center Buildout (arxiv.org)

Self-Guided LLM Reasoning for Error Diagnosis in Vehicle Routing Problems (arxiv.org)

Learning to Reason with Selective Internal Feedback (arxiv.org)

Hierarchical Graph Convolutional Network Approach for Detecting Low-Quality Documents (arxiv.org)

FedSkillGit: Federated Skill Evolution via Version Control (arxiv.org)

DST-Net: Decomposed Spatio-Temporal Network for Time Series Forecasting (arxiv.org)

Causal Syntax Circuits in CodeLMs: Budget and Probe Decide What Looks Localised (arxiv.org)

TurboMamba: Faster Inference, Faster Training, and Better Prediction with Lexical Memory (arxiv.org)

KNOU-Flood: Advancing global streamflow forecasting with lead-aware probabilistic machine learning (arxiv.org)

Same Answer, New Box: How Much of RLVR's Out-of-Domain Gain Is Answer Format? (arxiv.org)

MemCut: Every Secret Leaves a Trail in the Long-Term Memory of LLM Agents (arxiv.org)

ARGUS: An Agenda-Driven Framework for Reliable User Simulation (arxiv.org)

Quantum Versus Classical Machine Learning for Neuropsychiatric Disorders: A Feature-Matched Multi-Cohort Evaluation (arxiv.org)

DocPilot: Interactive Agent Orchestration and Training for Long Visual Document Understanding (arxiv.org)

MSP: Modality Self-Play for Training Multimodal LLMs Without Paired Cross-Modal Data (arxiv.org)

Online Quantile Omniprediction for Proper Losses (arxiv.org)

Does On-Policy Distillation Need to Complete Every Rollout? (arxiv.org)

Ranking and Partitioning of Attention Heads for Sparse and Continual Reinforcement Learning Fine-tuning (arxiv.org)

Feature Elicitation for SAE Steering of Deceptive Behaviour in Small LLMs (arxiv.org)

TimeCap: A Scalable Open-Source Pipeline for Captioning Wearable Health Time-Series Data (arxiv.org)

Follow the Perturbed Leader for Linear Bandits: $\sqrt{T}$ Regret over General Convex Sets (arxiv.org)

The Tilted Sampling Problem: Optimism, Posterior Sampling, and Logarithmic Regret (arxiv.org)

A Mixture-of-Experts Framework for Multi-Stream Quickest Change Detection (arxiv.org)

SelectPFN: Amortized Bayesian Variable Selection for Sparse Nonlinear Regression with Prior-Data Fitted Networks (arxiv.org)

Ranked Right, Sized Wrong: Ensemble Spread Fails Silently at Streamflow Extremes (arxiv.org)

Shortcut Sensitivity in Agent-Safety Evaluations (arxiv.org)

Graph Additive Models disentangling feature, topology, and interactions (arxiv.org)

Multi-task Neural Diffusion Processes (arxiv.org)

Parallel Layer Normalization for Universal Approximation (arxiv.org)

STAMP: Symmetric Temporal Averaging Message Passing for Domain Adaptation in Chronologically Split Graphs (arxiv.org)

A Closed Form Depth-Width Law for Neural Collapse in Deep Linear Networks (arxiv.org)

L2GSSL: Label-Efficient Local-to-Global Self-Supervised Learning for Volumetric Healthcare Data (arxiv.org)

Priming Base Models for Agentic Post-Training: An Open Midtraining Recipe (arxiv.org)

Single-stage joint diffusion of sky images and clear-sky index for ultra-short-term irradiance forecasting: what the frame-to-irradiance stage discards (arxiv.org)

Same Spectrum, Harder First-Order Training: Phase-Locking in Single- versus Multi-Qubit Re-Uploading Circuits (arxiv.org)

When Should a Low-Light Pipeline Generate? Auditable Risk Control for Selective Generative Enhancement (arxiv.org)