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。

Cut2Next: Generating Next Shot via In-Context Tuning

DIVER: A Multi-Stage Approach for Reasoning-intensive Information Retrieval

Omni-Effects: Unified and Spatially-Controllable Visual Effects Generation

Chimera: Harnessing Multi-Agent LLMs for Automatic Insider Threat Simulation

Klear-Reasoner: Advancing Reasoning Capability via Gradient-Preserving Clipping Policy Optimization

TurboBias: Universal ASR Context-Biasing powered by GPU-accelerated Phrase-Boosting Tree

AMFT: Aligning LLM Reasoners by Meta-Learning the Optimal Imitation-Exploration Balance

LSDT: LLM-Augmented Semantic Digital Twins for Adaptive Knowledge-Intensive Infrastructure Planning

Do Biased Models Have Biased Thoughts?

Early Detection of Pancreatic Cancer Using Multimodal Learning on Electronic Health Record

LLM Unlearning Without an Expert Curated Dataset

Multi-Faceted Large Embedding Tables for Pinterest Ads Ranking

Echo: Decoupling Inference and Training for Large-Scale RL Alignment on Heterogeneous Swarms

Situated Epistemic Infrastructures: A Diagnostic Framework for Post-Coherence Knowledge

RCR-Router: Efficient Role-Aware Context Routing for Multi-Agent LLM Systems with Structured Memory

Position: The Current AI Conference Model is Unsustainable! Diagnosing the Crisis of Centralized AI Conference

GTPO and GRPO-S: Token and Sequence-Level Reward Shaping with Policy Entropy

A Few Words Can Distort Graphs: Knowledge Poisoning Attacks on Graph-based Retrieval-Augmented Generation of Large Language Models

Explaining Time Series Classifiers with PHAR: Rule Extraction and Fusion from Post-hoc Attributions

Role-Aware Language Models for Secure and Contextualized Access Control in Organizations

DynaSwarm: Dynamically Graph Structure Selection for LLM ベースのマルチエージェントシステム

Post-Completion Learning for Language Models

Alternates, Assemble! Selecting Optimal Alternates for Citizens' Assemblies

Argus Inspection: Do Multimodal Large Language Models Possess the Eye of Panoptes?

RAGtifier: Evaluating RAG Generation Approaches of State-of-the-Art RAG Systems for the SIGIR LiveRAG Competition

Unsupervised Document and Template Clustering using Multimodal Embeddings

Saturation Self-Organizing Map

CulturalFrames: Assessing Cultural Expectation Alignment in Text-to-Image Models and Evaluation Metrics

To Judge or not to Judge: Using LLM Judgements for Advertiser Keyphrase Relevance at eBay

Edge-Cloud Collaborative Computing on Distributed Intelligence and Model Optimization: A Survey

Mj\"olnir: A Deep Learning Parametrization Framework for Global Lightning Flash Density

Federated Learning: A Survey on Privacy-Preserving Collaborative Intelligence

Democracy of AI Numerical Weather Models: An Example of Global Forecasting with FourCastNetv2 Made by a University Research Lab Using GPU

Retrieval-Augmented Generation with Conflicting Evidence

SPIE: Semantic and Structural Post-Training of Image Editing Diffusion Models with AI フィードバック

Evaluating Trust in AI, Human, and Co-produced Feedback Among Undergraduate Students

ProtoECGNet: Case-Based Interpretable Deep Learning for Multi-Label ECG Classification with Contrastive Learning

ChatBench: From Static Benchmarks to Human-AI Evaluation

Adaptive Computation Pruning for the Forgetting Transformer

AI-induced sexual harassment: Investigating Contextual Characteristics and User Reactions of Sexual Harassment by a Companion Chatbot

CrossWordBench: Evaluating the Reasoning Capabilities of LLMs and LVLMs with Controllable Puzzle Generation

Opioid Named Entity Recognition (ONER-2025) から Reddit

OSMa-Bench: Evaluating Open Semantic Mapping Under Varying Lighting Conditions

TIDE: Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation

Flexible Prefrontal Control over Hippocampal Episodic Memory for Goal-Directed Generalization

EvoP: Robust LLM Inference via Evolutionary Pruning

Sleepless Nights, Sugary Days: Creating Synthetic Users with Health Conditions for Realistic Coaching Agent Interactions

Zero-shot Emotion Annotation in Facial Images Using Large Multimodal Models: Benchmarking and Prospects for Multi-Class, Multi-Frame Approaches

PAR-AdvGAN: Improving Adversarial Attack Capability with Progressive Auto-Regression AdvGAN

Forget the Data and Fine-Tuning! Just Fold the Network to Compress

FBFL: A Field-Based Coordination Approach for Data Heterogeneity in Federated Learning

Decoding-based Regression

AdEval: Alignment-based Dynamic Evaluation to Mitigate Data Contamination in Large Language Models

Chemist-aligned retrosynthesis by ensembling diverse inductive bias models

Adaptive Informed Deep Neural Networks for Power Flow Analysis

A Risk Taxonomy and Reflection Tool for Large Language Model Adoption in Public Health

Learning Marmoset Vocal Patterns with a Masked Autoencoder for Robust Call Segmentation, Classification, and Caller Identification

Dynamic Spectrum Access for Ambient Backscatter Communication-assisted D2D Systems with Quantum Reinforcement Learning

Zero-Shot Generalization of Vision-Based RL Without Data Augmentation

Hypergraph-based Motion Generation with Multi-modal Interaction Relational Reasoning

3DFacePolicy: Audio-Driven 3D Facial Animation Based on Action Control

Return Prediction for Mean-Variance Portfolio Selection: How Decision-Focused Learning Shapes Forecasting Models

OE3DIS: Open-Ended 3D Point Cloud Instance Segmentation

VisionUnite: A Vision-Language Foundation Model for Ophthalmology Enhanced with Clinical Knowledge

DreamStory: Open-Domain Story Visualization by LLM-Guided Multi-Subject Consistent Diffusion

MEReQ: Max-Ent Residual-Q Inverse RL for Sample-Efficient Alignment from Intervention

Multidimensional Adaptive Coefficient for Inference Trajectory Optimization in Flow and Diffusion

AIOS: LLM Agent Operating System

Keep Your Friends Close: Leveraging Affinity Groups to Accelerate AI Inference Workflows

From Lab to Field: Real-World Evaluation of an AI-Driven Smart Video Solution to Enhance Community Safety

BELLA: Black box model Explanations by Local Linear Approximations

Artificial Intelligence Software Structured to Simulate Human Working Memory, Mental Imagery, and Mental Continuity

Fitting Description Logic Ontologies to ABox and Query Examples

Interpreting Fedspeak with Confidence: A LLM-Based Uncertainty-Aware Framework Guided by Monetary Policy Transmission Paths

Designing a Feedback-Driven Decision Support System for Dynamic Student Intervention

Large Language Models Do Not Simulate Human Psychology

IRL-VLA: Training an Vision-Language-Action Policy via Reward World Model

InfiAlign: A Scalable and Sample-Efficient Framework for Aligning LLMs to Enhance Reasoning Capabilities

SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience

Trainable Dynamic Mask Sparse Attention

Edge-Based Multimodal Sensor Data Fusion with Vision Language Models (VLMs) for Real-time Autonomous Vehicle Accident Avoidance

Cognitive Kernel-Pro: A Framework for Deep Research Agents and Agent Foundation Models Training

Probabilistic Active Goal Recognition

When Imitation Learning Outperforms Reinforcement Learning in Surgical Action Planning

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

UnrealZoo: Enriching Photo-realistic Virtual Worlds for Embodied AI

System~2 Reasoning for Human--AI Alignment: Generality and Adaptivity via ARC-AGI

Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models

Training-Free Text-Guided Color Editing with Multi-Modal Diffusion Transformer

Towards Universal Neural Inference

SPARC: Soft Probabilistic Adaptive multi-interest Retrieval Model via Codebooks for recommender system

Dynamic Uncertainty-aware Multimodal Fusion for Outdoor Health Monitoring

Can We Trust AI to Govern AI? Benchmarking LLM Performance on Privacy and AI Governance Exams

Spatial Traces: Enhancing VLA Models with Spatial-Temporal Understanding

E3-Rewrite: Learning to Rewrite SQL for Executability, Equivalence,and Efficiency

When Deepfakes Look Real: Detecting AI-Generated Faces with Unlabeled Data due to Annotation Challenges

Attacks and Defenses Against LLM Fingerprinting

LyS at SemEval 2025 Task 8: Zero-Shot Code Generation for Tabular QA

Retrospective Sparse Attention for Efficient Long-Context Generation

Rational Inverse Reasoning

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence

Created by

Haebom

作者

Hongzhe Du, Weikai Li, Min Cai, Karim Saraipour, Zimin Zhang, Himabindu Lakkaraju, Yizhou Sun, Shichang Zhang

概要

この論文は、事前訓練された大規模言語モデル（LLM）をより有用で整列した事後訓練モデルに変換するために不可欠な事後訓練プロセスがLLMの内部をどのように再構成するかについての機械的分析を提供します。モデル系列とデータセットにわたる4つの観点（事実上の知識格納場所、知識表現、真実性および拒否表現、信頼性）から、基本モデルと事後訓練モデルを比較分析して、次の結果を導き出しました。第一に、事後訓練は、現実的な知識保存場所を変更することなく、基本モデルの知識表現を適応させながら、新しい知識表現を開発する。第二に、真実性と拒否は隠された表現空間のベクトルとして表現することができ、真実性の方向は基本モデルと事後訓練モデルとの間で非常に似ており、介入に効果的に伝達されます。第三に、拒否方向は基本モデルと事後訓練モデルとで異なり、制限的な伝達性を示します。第四に、基本モデルと事後訓練モデルとの間の信頼性の違いは、エントロピーニューロンに起因することはできません。この研究は、ポストトレーニング中に維持および変更される基本的なメカニズムへの洞察を提供し、モデル調整などのフォローアップを容易にし、解析性およびLLMポストトレーニングの将来の研究に潜在的に役立つ可能性があります。

Takeaways、Limitations

•

Takeaways:

◦

事後訓練の基本メカニズムの理解を促進

◦

モデル調整などのフォローアップの改善に貢献

◦

LLM解析性と事後訓練研究の新しい方向性の提示

◦

事後訓練過程における知識表現の変化の様相

◦

真実性と拒否表現のベクトル表現可能性とその伝達性の分析

•

Limitations:

◦

特定のモデル系列とデータセットの分析結果なので、一般化に関するさらなる研究が必要

◦

エントロピーニューロン以外の他の要因による信頼度差の可能性を考慮する必要

◦

拒否方向の限られた伝達性のためのさらなる分析と改善方案の探求

Made with Slashpage