AI News

How Much Orthogonalization Does Muon Need? (arxiv.org)

CRMA: A Spectrally-Bounded Backbone for Modular Continual Fine-Tuning of LLMs (arxiv.org)

Detector-Evasive LLM Paraphrasing via Constrained Policy Optimization (arxiv.org)

PR2: Predictive Routing Replay for MoE-Based LLM Reinforcement Learning (arxiv.org)

Quantum Reservoir Computing and Risk Bounds (arxiv.org)

Subliminal Learning Is Steering Vector Distillation (arxiv.org)

Property Prediction of Stacked Bilayer Materials: A Multimodal Learning Approach (arxiv.org)

Can AI Review Improve Paper Drafting? An Empirical Study on 20 Computer Architecture Submissions (arxiv.org)

Tackling the Root of Misinformation by Teaching Laypeople about Logical Fallacies via Socratic Questioning and Critical Argumentation (arxiv.org)

TriLens: Per-Layer Logit-Lens Entropy for White-Box Hallucination Detection (arxiv.org)

Dynamic Proxy-Mixing: Transferring Replay Controllers from Small to Large Models for Continual Instruction Tuning (arxiv.org)

Auditing Near-Optimal Policies Can Be Exponentially Hard: Conditional Query Lower Bounds via Occupancy Rashomon Capacity (arxiv.org)

Finer Parameter Steps for Low-Rank PEFT: A Controlled Study with CP Tensor Adapters (arxiv.org)

Canonicalized Stable-List Replay for Private Federated Continual Learning over Language-Model Embeddings (arxiv.org)

VERA: Variational Inference Framework for Jailbreaking Large Language Models (arxiv.org)

Targeted Data Fusion for Region-Specific Survival Effects in the AMP HIV Prevention Trials (arxiv.org)

Non-vacuous Generalization Bounds for Deep Neural Networks without any modification to the trained models (arxiv.org)

Advancing Local Clustering on Graphs via Compressive Sensing: Semi-supervised and Unsupervised Methods (arxiv.org)

Global Convergence of Adaptive Sensing for Principal Eigenvector Estimation (arxiv.org)

Human in the Loop Adaptive Optimization for Improved Time Series Forecasting (arxiv.org)

Interpretable Self-Supervised Learning via Representer Landmarks and Nystr\"om Approximation (arxiv.org)

Trajectory Data Suffices for Statistically Efficient Policy Evaluation in Fixed-Horizon Offline RL with Linear $q^\pi$-Realizability and Concentrability (arxiv.org)

Safeguarded Stochastic Polyak Step Sizes for Non-smooth Optimization: Robust Performance Without Small (Sub)Gradients (arxiv.org)

Independent Component Discovery in Temporal Count Data (arxiv.org)

Universal One-third Time Scaling in Learning Peaked Distributions (arxiv.org)

Collaborative and Efficient Fine-tuning: Leveraging Task Similarity (arxiv.org)

WildCat: Near-Linear Attention in Theory and Practice (arxiv.org)

The Assistant as a Privileged Persona: A canonical reference in cross-persona self-recognition (arxiv.org)

Normalized Relevance Measure as a Unifying Framework to Explain Neural Network Latent Structures (arxiv.org)

Semi-Supervised Noise Adaptation: Transferring Knowledge from Noise Domain (arxiv.org)

Richer Representations for Neural Algorithmic Reasoning via Auxiliary Reconstruction (arxiv.org)

Interpretable Policy Distillation for Power Grid Topology Control (arxiv.org)

Same Payload, Different Channel: Measuring Trust Asymmetry in Tool-Using Language Models (arxiv.org)

On the Recoverability of Causal Relations from Bulk Gene Expression Data (arxiv.org)

On the Difficulty of Learning a Meta-network for Training Data Selection (arxiv.org)

Characterization of Multi-Model Agentic AI Systems on General Tasks via Trace-Driven Simulation (arxiv.org)

SIRIUS-SQL: Anchoring Multi-Candidate Text-to-SQL in Execution Feedback (arxiv.org)

ANDES: Agent Native Data Evolving Synthesis Tool for Autonomous Instruction Alignment (arxiv.org)

SkillSmith: Co-Evolving Skills and Tools for Self-Improving Agent Systems (arxiv.org)

Science Earth: Towards A Planet-Scale Operating System for AI-Native Scientific Discovery (arxiv.org)

Recognize Your Orchestrator: An Entropy Dynamics Perspective for LLM Multi-Agent Systems (arxiv.org)

FlowTime: Towards Continuous Generative Watch Time Prediction via Flow-based Personalized Priors (arxiv.org)

Spatiotemporal Multi-Task Graph Transformer for Trip-Level Transit Prediction (arxiv.org)

LASER: Loss-Aware Singular-value Decomposition and Rank Allocation for Efficient Low-Precision Vision-Language Models (arxiv.org)

CARE-RL: Capability-Aware Reinforcement Learning for Mitigating Cross-Domain Conflicts (arxiv.org)

How Neural Losses Shape VAE Latents (arxiv.org)

MESA: Improving MoE Safety Alignment via Decentralized Expertise (arxiv.org)

Early Diagnosis of Wasted Computation in Multi-Agent LLM Systems via Failure-Aware Observability (arxiv.org)

GuidaPA: Privacy-Preserving Chatbot for Public Administration via Federated Learning (arxiv.org)

Self-Healing Agentic Orchestrators for Reliable Tool-Augmented Large Language Model Systems (arxiv.org)