Daily Arxiv

전 세계에서 발간되는 인공지능 관련 논문을 정리하는 페이지 입니다.
본 페이지는 Google Gemini를 활용해 요약 정리하며, 비영리로 운영 됩니다.
논문에 대한 저작권은 저자 및 해당 기관에 있으며, 요약본 공유 시 출처만 명기하면 됩니다.
This service is supported by Google Gemini.

ENSEMBITS: an alphabet of protein conformational ensembles

(How) Do Large Language Models Understand High-Level Message Sequence Charts?

Query-Conditioned Test-Time Self-Training for Large Language Models

Constitutional Governance in Metric Spaces

Inducing Overthink: Hierarchical Genetic Algorithm-based DoS Attack on Black-Box Large Language Reasoning Models

LeanSearch v2: Global Premise Retrieval for Lean 4 Theorem Proving

MLGIB: Multi-Label Graph Information Bottleneck for Expressive and Robust Message Passing

Watermarking Should Be Treated as a Monitoring Primitive

Does language matter for spoken word classification? A multilingual generative meta-learning approach

Scaling few-shot spoken word classification with generative meta-continual learning

Context Training with Active Information Seeking

WriteSAE: Sparse Autoencoders for Recurrent State

Learning, Fast and Slow: Towards LLMs That Adapt Continually

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training

Detecting overfitting in Neural Networks during long-horizon grokking using Random Matrix Theory

A New Technique for AI Explainability using Feature Association Map

TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching

Boosting Omni-Modal Language Models: Staged Post-Training with Visually Debiased Evaluation

GEAR: Granularity-Adaptive Advantage Reweighting for LLM Agents via Self-Distillation

Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models

Predictive Maps of Multi-Agent Reasoning: A Successor-Representation Spectrum for LLM Communication Topologies

LoKA: Low-precision Kernel Applications for Recommendation Models At Scale

Prompt-Activation Duality: Improving Activation Steering via Attention-Level Interventions

Pretraining large language models with MXFP4 on Native FP4 Hardware

ShadowMerge: A Novel Poisoning Attack on Graph-Based Agent Memory via Relation-Channel Conflicts

Evolutionary Ensemble of Agents

Geometrically Constrained Stenosis Editing in Coronary Angiography via Entropic Optimal Transport

AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems

Trapping Attacker in Dilemma: Examining Internal Correlations and External Influences of Trigger for Defending GNN Backdoors

Normalization Equivariance for Arbitrary Backbones, with Application to Image Denoising

Flow-OPD: On-Policy Distillation for Flow Matching Models

One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy

Toward Privileged Foundation Models:LUPI for Accelerated and Improved Learning

Predict-then-Diffuse: Adaptive Response Length for Compute-Budgeted Inference in Diffusion LLMs

Motion-Aware Caching for Efficient Autoregressive Video Generation

ClawGym: A Scalable Framework for Building Effective Claw Agents

SIEVES: Selective Prediction Generalizes through Visual Evidence Scoring

Quantifying and Mitigating Self-Preference Bias of LLM Judges

Quotient-Space Diffusion Models

Image Generators are Generalist Vision Learners

RoboWM-Bench: A Benchmark for Evaluating World Models in Robotic Manipulation

Evaluating Adaptive Personalization of Educational Readings with Simulated Learners

UniMamba: A Unified Spatial-Temporal Modeling Framework with State-Space and Attention Integration

RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies

ECHO: Elastic Speculative Decoding with Sparse Gating for High-Concurrency Scenarios

PinpointQA: A Dataset and Benchmark for Small Object-Centric Spatial Understanding in Indoor Videos

LLMs Should Express Uncertainty Explicitly

A Security Analysis of the OpenClaw AI Agent Framework

OneSearch-V2: The Latent Reasoning Enhanced Self-distillation Generative Search Framework

Procedural Refinement by LLM-driven Algorithmic Debugging for ARC-AGI-2

Residual Stream Duality in Modern Transformer Architectures

M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling

Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages

TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning

Adapting Dijkstra for Buffers and Unlimited Transfers

Neural Field Thermal Tomography: A Differentiable Physics Framework for Non-Destructive Evaluation

V2M-Zero: Zero-Pair Time-Aligned Video-to-Music Generation

Gradient Iterated Temporal-Difference Learning

Vibe Code Bench: Evaluating AI Models on End-to-End Web Application Development

Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons

Decoupling Stability and Plasticity for Multi-Modal Test-Time Adaptation

MPU: Towards Secure and Privacy-Preserving Knowledge Unlearning for Large Language Models

Grokking Finite-Dimensional Algebra

CUICurate: A GraphRAG-based Framework for Automated Clinical Concept Curation for NLP applications

MALLVI: A Multi-Agent Framework for Integrated Generalized Robotics Manipulation

Artificial Intelligence Specialization in the European Union: Underexplored Role of the Periphery at NUTS-3 Level

Numerical exploration of the range of shape functionals using neural networks

Finding Interpretable Prompt-Specific Circuits in Language Models

ArGEnT: Arbitrary Geometry-encoded Transformer for Operator Learning

Krause Synchronization Transformers

Proximal Action Replacement for Behavior Cloning Actor-Critic in Offline Reinforcement Learning

VLRS-Bench: A Vision-Language Reasoning Benchmark for Remote Sensing

GhostCite: A Large-Scale Analysis of Citation Validity in the Age of Large Language Models

Boosting LLM Reasoning via Human-Inspired Reward Shaping

PaAno: Patch-Based Representation Learning for Time-Series Anomaly Detection

Neural Signals Generate Clinical Notes in the Wild

L2R: Low-Rank and Lipschitz-Controlled Routing for Mixture-of-Experts

OPT-Engine: Benchmarking the Limits of LLMs in Optimization Modeling via Complexity Scaling

Teaching and Evaluating LLMs to Reason About Polymer Design Related Tasks

The Multi-View Paradigm Shift in MRI Radiomics: Predicting MGMT Methylation in Glioblastoma

LangPrecip: Language-Aware Multimodal Precipitation Nowcasting

Achieving Approximate Symmetry Is Exponentially Easier than Exact Symmetry

Why Goal-Conditioned Reinforcement Learning Works: Relation to Dual Control

AI-Driven Optimization under Uncertainty for Mineral Processing Operations

fMRI-LM: Towards a Universal Foundation Model for Language-Aligned fMRI Understanding

A cross-species neural foundation model for end-to-end speech decoding

How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining

Optimizing PyTorch Inference with LLM-Based Multi-Agent Systems

Chinese Short-Form Creative Content Generation via Explanation-Oriented Multi-Objective Optimization

Descriptor: Distance-Annotated Traffic Perception Question Answering (DTPQA)

Moral Susceptibility and Robustness under Persona Role-Play in Large Language Models

From Ranking to Reasoning: Explainable Web API Recommendation via Semantic Reasoning

Higher-order Linear Attention

A Tutorial on Cognitive Biases in Agentic AI-Driven 6G Autonomous Networks

Seeing Through the Brain: New Insights from Decoding Visual Stimuli with fMRI

AMiD: Knowledge Distillation for LLMs with $\alpha$-mixture Assistant Distribution

VER: Vision Expert Transformer for Robot Learning via Foundation Distillation and Dynamic Routing

TiTok: Transfer Token-level Knowledge via Contrastive Excess to Transplant LoRA

The Pitfalls of KV Cache Compression

AVEX: What Matters for Animal Vocalization Encoding

Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective

Created by

Haebom

저자

Yang Zhang, Xinran Li, Jianing Ye, Delin Qu, Shuang Qiu, Chongjie Zhang, Xiu Li, Chenjia Bai

개요

본 논문은 다중 에이전트 강화 학습(MARL)에서 세계 모델링의 복잡성을 줄이기 위해 순차적 에이전트 모델링을 제안합니다. 기존의 전체 상태-행동 전이 역학을 공동으로 모델링하는 대신, 각 시간 단계에서 상태 공간에만 집중하여 모델링 복잡성을 줄입니다. 이는 에이전트 간의 구조적 의존성을 포착하면서 불확실성을 점진적으로 해소하여 에이전트가 상태에 미치는 영향을 더 정확하게 나타냅니다. 이러한 순차적인 에이전트 행동 공개 방식은 확산 모델의 역과정과 유사하며, 이를 활용하여 확산 모델 기반의 유연하고 강력한 MARL 세계 모델인 DIMA를 개발합니다. DIMA는 MAMuJoCo와 Bi-DexHands 등 여러 다중 에이전트 제어 벤치마크에서 최첨단 성능을 달성하며, 기존 세계 모델보다 최종 보상과 샘플 효율성 면에서 뛰어난 성능을 보입니다.