Daily Arxiv

世界中で発行される人工知能関連の論文をまとめるページです。
このページはGoogle Geminiを活用して要約し、非営利で運営しています。
論文の著作権は著者および関連機関にあり、共有する際は出典を明記してください。

Dense Video Understanding with Gated Residual Tokenization

Machines are more productive than humans until they aren't, and vice versa

BWCache: Accelerating Video Diffusion Transformers through Block-Wise Caching

Exploring Data and Parameter Efficient Strategies for Arabic Dialect Identifications

The threat of analytic flexibility in using large language models to simulate human data: A call to attention

Evaluating undergraduate mathematics examinations in the era of generative AI: a curriculum-level case study

A Graph-Based Approach to Alert Contextualisation in Security Operations Centres

FunAudio-ASR Technical Report

Omni-CLST: Error-aware Curriculum Learning with guided Selective chain-of-Thought for audio question answering

Do Code Semantics Help? A Comprehensive Study on Execution Trace-Based Information for Code Large Language Models

Pluralistic Alignment for Healthcare: A Role-Driven Framework

ALIGNS: Unlocking nomological networks in psychological measurement through a large language model

A Survey of Reinforcement Learning for Large Reasoning Models

Skeleton-based sign language recognition using a dual-stream spatio-temporal dynamic graph convolutional network

Reconstruction Alignment Improves Unified Multimodal Models

Moment- and Power-Spectrum-Based Gaussianity Regularization for Text-to-Image Models

FASL-Seg: Anatomy and Tool Segmentation of Surgical Scenes

Dual-Mode Deep Anomaly Detection for Medical Manufacturing: Structural Similarity and Feature Distance

Exploit Tool Invocation Prompt for Tool Behavior Hijacking in LLM-Based Agentic System

Measuring the Measures: Discriminative Capacity of Representational Similarity Metrics Across Model Families

AR-KAN: Autoregressive-Weight-Enhanced Kolmogorov-Arnold Network for Time Series Forecasting

Ensemble of Pathology Foundation Models for MIDOG 2025 Track 2: Atypical Mitosis Classification

Deep Learning-Driven Multimodal Detection and Movement Analysis of Objects in Culinary

Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning

MovieCORE: COgnitive REasoning in Movies

ASE: A Repository-Level Benchmark for Evaluating Security in AI-Generated Code

Generalized invariants meet constitutive neural networks: A novel framework for hyperelastic materials

Neural Logic Networks for Interpretable Classification

Roll Your Eyes: Gaze Redirection via Explicit 3D Eyeball Rotation

Controllable Surface Diffusion Generative Model for Neurodevelopmental Trajectories

Deciding how to respond: A deliberative framework to guide policymaker responses to AI systems

SCORPION: Addressing Scanner-Induced Variability in Histopathology

ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning

SPICE: An Automated SWE-Bench Labeling Pipeline for Issue Clarity, Test Coverage, and Effort Estimation

FreeAudio: Training-Free Timing Planning for Controllable Long-Form Text-to-Audio Generation

EnCoBo: Energy-Guided Concept Bottlenecks for Interpretable Generation

T-SYNTH: A Knowledge-Based Dataset of Synthetic Breast Images

MedVAL: Toward Expert-Level Medical Text Validation with Language Models

Survivability of Backdoor Attacks on Unconstrained Face Recognition Systems

"What's Up, Doc?": Analyzing How Users Seek Health Information in Large-Scale Conversational AI Datasets

Engineering RAG Systems for Real-World Applications: Design, Development, and Evaluation

An Explainable AI Framework for Dynamic Resource Management in Vehicular Network Slicing

DiCoRe: Enhancing Zero-shot Event Detection via Divergent-Convergent LLM Reasoning

Semantic Exploration and Dense Mapping of Complex Environments using Ground Robot with Panoramic LiDAR-Camera Fusion

Evaluating Supervised Learning Models for Fraud Detection: A Comparative Study of Classical and Deep Architectures on Imbalanced Transaction Data

Binarized Neural Networks Converge Toward Algorithmic Simplicity: Empirical Support for the Learning-as-Compression Hypothesis

PMPO: Probabilistic Metric Prompt Optimization for Small and Large Language Models

DisastIR: A Comprehensive Information Retrieval Benchmark for Disaster Management

Preference Isolation Forest for Structure-based Anomaly Detection

Trustless Autonomy: Understanding Motivations, Benefits, and Governance Dilemmas in Self-Sovereign Decentralized AI Agents

GRADA: Graph-based Reranking against Adversarial Documents Attack

Modular Machine Learning: An Indispensable Path towards New-Generation Large Language Models

Direct Video-Based Spatiotemporal Deep Learning for Cattle Lameness Detection

Read Before You Think: Mitigating LLM Comprehension Failures with Step-by-Step Reading

Zero-Shot LLMs in Human-in-the-Loop RL: Replacing Human Feedback for Reward Shaping

Predicting Multi-Agent Specialization via Task Parallelizability

Fine-tuning Vision Language Models with Graph-based Knowledge for Explainable Medical Image Analysis

VLM-E2E: Enhancing End-to-End Autonomous Driving with Multimodal Driver Attention Fusion

METAL: A Multi-Agent Framework for Chart Generation with Test-Time Scaling

SNaRe: Domain-aware Data Generation for Low-Resource Event Detection

Superpose Task-specific Features for Model Merging

Examining False Positives under Inference Scaling for Mathematical Reasoning

SWAT: Sliding Window Adversarial Training for Gradual Domain Adaptation

Advanced Physics-Informed Neural Network with Residuals for Solving Complex Integral Equations

Retrieval-Retro: Retrieval-based Inorganic Retrosynthesis with Expert Knowledge

Unlocking Legal Knowledge: A Multilingual Dataset for Judicial Summarization in Switzerland

Reconstruction of Differentially Private Text Sanitization via Large Language Models

3DS: Medical Domain Adaptation of LLMs via Decomposed Difficulty-based Data Selection

The Role of Graph Topology in the Performance of Biomedical Knowledge Graph Completion Models

Top K Enhanced Reinforcement Learning Attacks on Heterogeneous Graph Node Classification

Learn while Unlearn: An Iterative Unlearning Framework for Generative Language Models

EXPLOR: Extrapolatory Pseudo-Label Matching for Out-of-distribution Uncertainty Based Rejection

Spatio-Temporal Anomaly Detection with Graph Networks for Data Quality Monitoring of the Hadron Calorimeter

Rule-Based Error Detection and Correction to Operationalize Movement Trajectory Classification

Heterogeneous Directed Hypergraph Neural Network over Abtract syntax tree (AST) for Code Classification

The Art of Saying "Maybe": A Conformal Lens for Uncertainty Benchmarking in VLMs

Human + AI for Accelerating Ad Localization Evaluation

Statistical Methods in Generative AI

InMind: Evaluating LLMs in Capturing and Applying Individual Human Reasoning Styles

DSperse: A Framework for Targeted Verification in Zero-Knowledge Machine Learning

DualSG: A Dual-Stream Explicit Semantic-Guided Multivariate Time Series Forecasting Framework

Judging with Many Minds: Do More Perspectives Mean Less Prejudice? On Bias Amplifications and Resistance in Multi-Agent Based LLM-as-Judge

Mastering Multi-Drone Volleyball through Hierarchical Co-Self-Play Reinforcement Learning

Automatic Mapping of AutomationML Files to Ontologies for Graph Queries and Validation

Explicit Context-Driven Neural Acoustic Modeling for High-Fidelity RIR Generation

FlowRL: Matching Reward Distributions for LLM Reasoning

Orion: Fuzzing Workflow Automation

TITAN: A Trajectory-Informed Technique for Adaptive Parameter Freezing in Large-Scale VQE

Fast and Fluent Diffusion Language Models via Convolutional Decoding and Rejective Fine-tuning

SMARTER: A Data-efficient Framework to Improve Toxicity Detection with Explanation via Self-augmenting Large Language Models

Watermarking and Anomaly Detection in Machine Learning Models for LORA RF Fingerprinting

Semi-Supervised 3D Medical Segmentation from 2D Natural Images Pretrained Model

Leveraging Geometric Visual Illusions as Perceptual Inductive Biases for Vision Models

Exploring How Audio Effects Alter Emotion with Foundation Models

WorldForge: Unlocking Emergent 3D/4D Generation in Video Diffusion Model via Training-Free Guidance

The mechanization of science illustrated by the Lean formalization of the multi-graded Proj construction

Vulnerable Agent Identification in Large-Scale Multi-Agent Reinforcement Learning

TextMine: LLM-Powered Knowledge Extraction for Humanitarian Mine Action

Listening, Imagining \& Refining: A Heuristic Optimized ASR Correction Framework with LLMs

Communication Efficient Split Learning of ViTs with Attention-based Double Compression

FlowRL: Matching Reward Distributions for LLM Reasoning

Created by

Haebom

作者

Xuekai Zhu, Daixuan Cheng, Dinghuai Zhang, Hengli Li, Kaiyan Zhang, Che Jiang, Youbang Sun, Ermo Hua, Yuxin Zuo, Xingtai Lv, Qizheng Zhang, Lin Chen, Fanghao Shao, Nong, Zing Jianfeng Gao, Xiaodong Liu, Bowen Zhou, Hongyuan Mei, Zhouhan Lin

概要

本稿では、大規模言語モデル（LLM）強化学習（RL）で報酬を最大化するのではなく、フローバランスを介して補償分布全体を一致させるFlowRLを提案します。最近開発された推論モデルは、PPOやGRPOなどの補償を最大化する方法を採用しています。対照的に、本論文では、学習可能な分割関数を使用してスカラー補償を正規化された目標分布に変換し、ポリシーと目標分布の間の逆KL発散を最小限に抑えます。このアイデアは、さまざまなナビゲーションと一般化可能な推論経路を促進するフローバランス最適化方法で実装されています。数学とコード推論作業の実験を行った結果、FlowRLは数学ベンチマークでGRPOより平均10.0％、PPOより5.1％向上し、コード推論作業でも一貫して優れたパフォーマンスを示しました。これらの結果は、LLM強化学習における効率的な探索と多様な推論のための重要なステップとして、報酬分布の一致を強調します。

Takeaways、Limitations

•

Takeaways：

◦

LLM強化学習における報酬の最大化の代わりに報酬分布マッチングの有効性を提示する。

◦

FlowRLが従来の方法（PPO、GRPO）よりも数学およびコード推論操作でより良いパフォーマンスを示すことを実験的に証明します。

◦

様々な推論経路の探索と一般化可能な推論経路の生成に貢献する。

•

Limitations：

◦

提示された方法の一般化性能のさらなる検証が必要である。

◦

特定のタスク（数学とコード推論）の実験結果のみが提示され、他のタスクへの適用可能性にはさらなる研究が必要です。

◦

学習可能な分割関数の設計と最適化の詳細な説明が不足している可能性があります。

Made with Slashpage