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。

Merge-of-Thought Distillation

OTESGN: Optimal Transport-Enhanced Syntactic-Semantic Graph Networks for Aspect-Based Sentiment Analysis

MESH - Understanding Videos Like Human: Measuring Hallucinations in Large Video Models

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

Symmetry-Guided Multi-Agent Inverse Reinforcement Learning

AU-Harness: An Open-Source Toolkit for Holistic Evaluation of Audio LLMs

Expert-Guided Explainable Few-Shot Learning for Medical Image Diagnosis

Towards Generalized Routing: Model and Agent Orchestration for Adaptive and Efficient Inference

MachineLearningLM: Scaling Many-shot In-context Learning via Continued Pretraining

Demo: Healthcare Agent Orchestrator (HAO) for Patient Summarization in Molecular Tumor Boards

Focusing by Contrastive Attention: Enhancing VLMs' Visual Reasoning

Beyond the Pre-Service Horizon: Infusing In-Service Behavior for Improved Financial Risk Forecasting

On Synthesis of Timed Regular Expressions

TinyDef-DETR: A DETR-based Framework for Defect Detection in Transmission Lines from UAV Imagery

LiDAR-BIND-T: Improved and Temporally Consistent Sensor Modality Translation and Fusion for Robotic Applications

From Vision to Validation: A Theory- and Data-Driven Construction of a GCC-Specific AI Adoption Index

A Comprehensive Guide to Differential Privacy: From Theory to User Expectations

The Architecture of AI Transformation: Four Strategic Patterns and an Emerging Frontier

FLM-Audio: Natural Monologues Improves Native Full-Duplex Chatbots via Dual Training

Deep Learning-Based Rock Particulate Classification Using Attention-Enhanced ConvNeXt

The Information Dynamics of Generative Diffusion

Data-Augmented Few-Shot Neural Stencil Emulation for System Identification of Computer Models

Group Expectation Policy Optimization for Heterogeneous Reinforcement Learning

Pretrained Conformers for Audio Fingerprinting and Retrieval

Towards Scalable Training for Handwritten Mathematical Expression Recognition

To Theoretically Understand Transformer-Based In-Context Learning for Optimizing CSMA

Klear-CodeTest: Scalable Test Case Generation for Code Reinforcement Learning

HiD-VAE: Interpretable Generative Recommendation via Hierarchical and Disentangled Semantic IDs

MagicGUI: A Foundational Mobile GUI Agent with Scalable Data Pipeline and Reinforcement Fine-tuning

Villa-X: Enhancing Latent Action Modeling in Vision-Language-Action Models

New Kid in the Classroom: Exploring Student Perceptions of AI Coding Assistants

Can Large Language Models Understand As Well As Apply Patent Regulations to Pass a Hands-On Patent Attorney Test?

Uncertainty-aware Diffusion and Reinforcement Learning for Joint Plane Localization and Anomaly Diagnosis in 3D Ultrasound

Uncertainty Estimation by Human Perception versus Neural Models

Persistent Homology of Topic Networks for the Prediction of Reader Curiosity

Task Matters: Knowledge Requirements Shape LLM Responses to Context-Memory Conflict

Crack Path Prediction with Operator Learning using Discrete Particle System data Generation

Diffusion Graph Neural Networks for Robustness in Olfaction Sensors and Datasets

MM-Prompt: Cross-Modal Prompt Tuning for Continual Visual Question Answering

An Ontology-Driven Graph RAG for Legal Norms: A Structural, Temporal, and Deterministic Approach

Combating Falsification of Speech Videos with Live Optical Signatures (Extended Version)

Early Exit and Multi Stage Knowledge Distillation in VLMs for Video Summarization

Critical Challenges and Guidelines in Evaluating Synthetic Tabular Data: A Systematic Review

Parasite: A Steganography-based Backdoor Attack Framework for Diffusion Models

Towards Adaptive Memory-Based Optimization for Enhanced Retrieval-Augmented Generation

Entropy-Gated Branching for Efficient Test-Time Reasoning

SWI: Speaking with Intent in Large Language Models

Byzantine-Robust Federated Learning Using Generative Adversarial Networks

VeriSafe Agent: Safeguarding Mobile GUI Agent via Logic-based Action Verification

MIND: Towards Immersive Psychological Healing with Multi-agent Inner Dialogue

V-HOP: Visuo-Haptic 6D Object Pose Tracking

EgoAgent: A Joint Predictive Agent Model in Egocentric Worlds

Knowledge-Guided Biomarker Identification for Label-Free Single-Cell RNA-Seq Data: A Reinforcement Learning Perspective

MERaLiON-SpeechEncoder: Towards a Speech Foundation Model for Singapore and Beyond

RED: Unleashing Token-Level Rewards from Holistic Feedback via Reward Redistribution

IDEATOR: Jailbreaking and Benchmarking Large Vision-Language Models Using Themselves

DeepVoting: Learning and Fine-Tuning Voting Rules with Canonical Embeddings

Rethinking Disentanglement under Dependent Factors of Variation

Discovering physical laws with parallel symbolic enumeration

Semantic Augmentation in Images using Language

Algorithmic Collusion by Large Language Models

A minimal coalition logic

Deep Reinforcement Learning for Inventory Networks: Toward Reliable Policy Optimization

Inconsistency Handling in Prioritized Databases with Universal Constraints: Complexity Analysis and Links with Active Integrity Constraints

Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference

CogGuide: Human-Like Guidance for Zero-Shot Omni-Modal Reasoning

TreeGPT: Pure TreeFFN Encoder-Decoder Architecture for Structured Reasoning Without Attention Mechanisms

Robix: A Unified Model for Robot Interaction, Reasoning and Planning

KROMA: Ontology Matching with Knowledge Retrieval and Large Language Models

Scaling LLM Planning: NL2FLOW for Parametric Problem Generation and Rigorous Evaluation

Optimizing Length Compression in Large Reasoning Models

LLMs for sensory-motor control: Combining in-context and iterative learning

Effort-aware Fairness: Incorporating a Philosophy-informed, Human-centered Notion of Effort into Algorithmic Fairness Metrics

Simulating Human-like Daily Activities with Desire-driven Autonomy

Enhancing Few-Shot Transfer Learning with Optimized Multi-Task Prompt Tuning through Modular Prompt Composition

ButterflyQuant: Ultra-low-bit LLM Quantization through Learnable Orthogonal Butterfly Transforms

CDE: Curiosity-Driven Exploration for Efficient Reinforcement Learning in Large Language Models

SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

Feasibility-Guided Fair Adaptive Offline Reinforcement Learning for Medicaid Care Management

Retrieval-Augmented Generation for Reliable Interpretation of Radio Regulations

Explaining Concept Drift through the Evolution of Group Counterfactuals

LoCoBench: A Benchmark for Long-Context Large Language Models in Complex Software Engineering

Mechanistic Learning with Guided Diffusion Models to Predict Spatio-Temporal Brain Tumor Growth

Graph Alignment via Dual-Pass Spectral Encoding and Latent Space Communication

ObjectReact: Learning Object-Relative Control for Visual Navigation

Fluent but Unfeeling: The Emotional Blind Spots of Language Models

Invisible Attributes, Visible Biases: Exploring Demographic Shortcuts in MRI-based Alzheimer's Disease Classification

An improved educational competition optimizer with multi-covariance learning operators for global optimization problems

Improving Video Diffusion Transformer Training by Multi-Feature Fusion and Alignment from Self-Supervised Vision Encoders

A modified RIME algorithm with covariance learning and diversity enhancement for numerical optimization

Towards Explainable Job Title Matching: Leveraging Semantic Textual Relatedness and Knowledge Graphs

Explainable AI for Accelerated Microstructure Imaging: A SHAP-Guided Protocol on the Connectome 2.0 scanner

Incorporating AI Incident Reporting into Telecommunications Law and Policy: Insights from India

OpenFake: An Open Dataset and Platform Toward Large-Scale Deepfake Detection

Prompt Pirates Need a Map: Stealing Seeds helps Stealing Prompts

Resource-Efficient Glioma Segmentation on Sub-Saharan MRI

ENSI: Efficient Non-Interactive Secure Inference for Large Language Models

We're Still Doing It (All) Wrong: Recommender Systems, Fifteen Years Later

LLMs Don't Know Their Own Decision Boundaries: The Unreliability of Self-Generated Counterfactual Explanations

MetaLLMix : An XAI Aided LLM-Meta-learning Based Approach for Hyper-parameters Optimization

RED: Unleashing Token-Level Rewards from Holistic Feedback via Reward Redistribution

Created by

Haebom

作者

Jiahui Li, Lin Li, Tai-wei Chang, Kun Kuang, Long Chen, Jun Zhou, Cheng Yang

概要

この論文では、人間のフィードバックからの強化学習（RLHF）を使用して、大規模言語モデル（LLM）を人間の好みに合わせる方法を提案します。従来のＲＬＨＦは、出力シーケンス全体に単一の補償を与える方式であり、個々のトークンの寄与を考慮に入れない限界がある。これを解決するために、本稿では、既存の報酬モデルを活用して各トークンの報酬を再配布する新しい方法であるRED（Reward Redistribution）を提示します。 REDは、既存の報酬モデルを変更したり、追加の学習ステップを必要とせずに、トークン単位の細分化された報酬を提供してLLMのパフォーマンスを向上させます。さまざまなデータセットと操作の実験結果は、REDの卓越性を示しています。

Takeaways、Limitations

•

Takeaways：

◦

既存のRLHFのLimitationsである単一補償方式を改善し、LLMの性能向上を図ります。

◦

トークン単位のきめ細かな補償により、LLMの言語的ニュアンスの理解を高めます。

◦

既存の補償モデルを変更することなく効果的なパフォーマンス向上を実現し、計算コストを最小限に抑えます。

◦

さまざまなデータセットやタスクで優れたパフォーマンスを実証します。

•

Limitations：

◦

提示されたRED法の一般化性能のさらなる研究が必要である。

◦

特定の種類の補償モデルにのみ適用可能です。

◦

トークンレベルの報酬分配方式の最適化に関する追加の研究が必要になる場合がある。

Made with Slashpage