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.

A New Technique for AI Explainability using Feature Association Map

Hide to See: Reasoning-prefix Masking for Visual-anchored Thinking in VLM Distillation

Gradient-Free Noise Optimization for Reward Alignment in Generative Models

LiBaGS: Lightweight Boundary Gap Synthesis for Targeted Synthetic Data Selection

HEPA: A Self-Supervised Horizon-Conditioned Event Predictive Architecture for Time Series

MCPShield: Content-Aware Attack Detection for LLM Agent Tool-Call Traffic

TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment

DataMaster: Data-Centric Autonomous AI Research

Shields to Guarantee Probabilistic Safety in MDPs

ALAM: Algebraically Consistent Latent Action Model for Vision-Language-Action Models

CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Driving

Metis: Learning to Jailbreak LLMs via Self-Evolving Metacognitive Policy Optimization

Combining Mechanical and Agentic Specification Inference for Move

Key-Value Means: Transformers with Expandable Block-Recurrent Compressed Memory

Pretraining large language models with MXFP4 on Native FP4 Hardware

FactoryNet: A Large-Scale Dataset toward Industrial Time-Series Foundation Models

When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models

Does Your Neural Network Extrapolate? Feature Engineering as Identifiability Bias for OOD Generalization

MathlibPR: Pull Request Merge-Readiness Benchmark for Formal Mathematical Libraries

AI CFD Scientist: Toward Open-Ended Computational Fluid Dynamics Discovery with Physics-Aware AI Agents

Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level

Multi-Dimensional Behavioral Evaluation of Agentic Stock Prediction Systems Using Large Language Model Judges with Closed-Loop Reinforcement Learning Feedback

Gyan: An Explainable Neuro-Symbolic Language Model

Efficient Geometry-Controlled High-Resolution Satellite Image Synthesis

Ilov3Splat: Instance-Level Open-Vocabulary 3D Scene Understanding in Gaussian Splatting

Talk is Cheap, Communication is Hard: Dynamic Grounding Failures and Repair in Multi-Agent Negotiation

COHERENCE: Benchmarking Fine-Grained Image-Text Alignment in Interleaved Multimodal Contexts

AgenticRecTune: Multi-Agent with Self-Evolving Skillhub for Recommendation System Optimization

Evaluation of Prompt Injection Defenses in Large Language Models

AmaraSpatial-10K: A Spatially and Semantically Aligned 3D Dataset for Spatial Computing and Embodied AI

Institutions for the Post-Scarcity of Judgment

Is SAM3 ready for pathology segmentation?

TStore: Rethinking AI Model Hub with Tensor-Centric Compression

(Sparse) Attention to the Details: Preserving Spectral Fidelity in ML-based Weather Forecasting Models

Technically Love: The Evolution of Human-AI Romance Discourse on Reddit

Skill-Conditioned Visual Geolocation for Vision-Language Models

LINE: LLM-based Iterative Neuron Explanations for Vision Models

Is a Picture Worth a Thousand Words? Adaptive Multimodal Fact-Checking with Visual Evidence Necessity

AgenticFlict: A Large-Scale Dataset of Merge Conflicts in AI Coding Agent Pull Requests on GitHub

Filter-then-Weight: Online Data Selection and Reweighting for LLM Fine-Tuning

MCLR: Improving Conditional Modeling via Inter-Class Likelihood-Ratio Maximization and Unifying Classifier-Free Guidance with Alignment Objectives

TiCo: Time-Controllable Spoken Dialogue Model

Delightful Distributed Policy Gradient

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

FlashSampling: Fast and Memory-Efficient Exact Sampling

A Benchmark for Multi-Party Negotiation Games from Real Negotiation Data

From Baselines to Transport Geodesics: Axiomatic Attribution via Optimal Generative Flows

Language Model Goal Selection Differs from Humans' in a Self-Directed Learning Task

Preconditioned Flow Matching

Kiwi-Edit: Versatile Video Editing via Instruction and Reference Guidance

Decentralized Ranking Aggregation via Gossip: Convergence and Robustness

Zatom-1: Towards a Multimodal Foundation Model for 3D Molecules and Materials

Test-Time Training with KV Binding Is Secretly Linear Attention

interwhen: A Generalizable Framework for Steering Reasoning Models with Test-time Verification

Entropy Aware Reward Guidance for Diffusion Language Model Alignment

Exponential Approximation Rates and Parameter Efficiency of Learnable Bernstein Activations

Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding

Auditing Sybil: Explaining Deep Lung Cancer Risk Prediction Through Generative Interventional Attributions

Preserve-Then-Quantize: Balancing Rank Budgets for Quantization Error Reconstruction in LLMs

RAG-GNN: Integrating Retrieved Knowledge with Graph Neural Networks for Precision Medicine

Optimization, Generalization and Differential Privacy Bounds for Gradient Descent on Kolmogorov-Arnold Networks

Improving Classifier-Free Guidance of Flow Matching via Manifold Projection

SynCABEL: Synthetic Contextualized Augmentation for Biomedical Entity Linking

GUIGuard-Bench: Toward a General Evaluation for Privacy-Preserving GUI Agents

High-Rate Quantized Matrix Multiplication I

Automated Rubrics for Reliable Evaluation of Medical Dialogue Systems

FlowPlan-G2P: A Structured Generation Framework for Transforming Scientific Papers into Patent Descriptions

Deep Delta Learning

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

Feature Learning Dynamics in Infinite-Depth Neural Networks

Generative Modeling from Black-box Corruptions via Self-Consistent Stochastic Interpolants

FOAM: Blocked State Folding for Memory-Efficient LLM Training

StreamGaze: Gaze-Guided Temporal Reasoning and Proactive Understanding in Streaming Videos

When Diffusion Breaks Constraints: Sequential Autoregressive Generation with RL and MCTS

SpectralTrain: A Universal Framework for Hyperspectral Image Classification

Radial Compensation: Fixing Radius Distortion in Chart-Based Generative Models on Riemannian Manifolds

CodeClash: Benchmarking Goal-Oriented Software Engineering

Higher-order Linear Attention

On the Sample Complexity of Differentially Private Policy Optimization

Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs

Latent-Augmented Discrete Diffusion Models

VERA-MH Concept Paper

Reinforcement Learning for Unsupervised Domain Adaptation in Spatio-Temporal Echocardiography Segmentation

REAP the Experts: Why Pruning Prevails for One-Shot MoE compression

Semantic knowledge guides innovation and drives cultural evolution

UniJEPA: Enhancing Robot Policy via Unified Continuous and Discrete Representation Learning

The Bayesian Origin of the Probability Weighting Function in Human Representation of Probabilities

Quantitative Certification of Agentic Tool Selection

Characteristic Root Analysis and Regularization for Linear Time Series Forecasting

DAWM: Diffusion Action World Models for Offline Reinforcement Learning via Action-Inferred Transitions

Adapting Vision-Language Models for Neutrino Event Classification in High-Energy Physics

GLASS: Global-Local Aggregation for Inference-time Sparsification of LLMs

Exact Verification of Graph Neural Networks with Incremental Constraint Solving

A Markov Categorical Framework for Language Modeling

RDMA: Cost Effective Agent-Driven Rare Disease Mining from Electronic Health Records

Where Do Reasoning Models Refuse?

Robotic Manipulation by Imitating Generated Videos Without Physical Demonstrations

LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing

Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models

T-TExTS (Teaching Text Expansion for Teacher Scaffolding): Enhancing Text Selection in High School Literature through Knowledge Graph-Based Recommendation

An Automated Reinforcement Learning Reward Design Framework with Large Language Model for Cooperative Platoon Coordination

Created by

Haebom

저자

Dixiao Wei, Peng Yi, Jinlong Lei, Yiguang Hong, Yuchuan Du

개요

본 논문은 강화학습(RL) 기반의 플래툰 조정 문제에서 성능이 좋은 보상 함수를 자동으로 생성하는 방법을 제시합니다. 기존 RL 기반 플래툰 조정은 수동으로 보상 함수를 설계해야 하는 어려움이 있었는데, 이 논문에서는 플래툰 조정 보상 설계 문제(PCRDP)를 정의하고, 대규모 언어 모델(LLM) 기반의 플래툰 조정 보상 설계(PCRD) 프레임워크를 제안합니다. PCRD는 분석 및 초기 보상(AIR) 모듈과 진화 모듈로 구성되어 있으며, LLM을 이용하여 환경 코드와 작업 요구 사항을 기반으로 보상 함수를 초기화하고, 훈련 피드백을 통해 반복적으로 최적화합니다. AIR 모듈은 사고의 연쇄를 통해 LLM의 코드 및 작업 이해도를 높이고, 진화 모듈은 탐색 다양성과 수렴 안정성 사이의 균형을 맞춥니다. 양쯔강 삼각주 교통망 시뮬레이션 내 6가지 복잡한 조정 시나리오를 통해 실험을 진행한 결과, PCRD로 생성된 보상 함수를 사용한 RL 에이전트가 기존 수동 설계 보상 함수보다 평균 10% 높은 성능을 달성했습니다.