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, 공유 시 출처만 명기하면 됩니다.

Instella: Fully Open Language Models with Stellar Performance

STAGE: A Symbolic Tensor grAph GEnerator for distributed AI system co-design

Speech-Audio Compositional Attacks on Multimodal LLMs and Their Mitigation with SALMONN-Guard

PustakAI: Curriculum-Aligned and Interactive Textbooks Using Large Language Models

Harnessing Bounded-Support Evolution Strategies for Policy Refinement

Potent but Stealthy: Rethink Profile Pollution against Sequential Recommendation via Bi-level Constrained Reinforcement Paradigm

FAST-CAD: A Fairness-Aware Framework for Non-Contact Stroke Diagnosis

Learning Topology-Driven Multi-Subspace Fusion for Grassmannian Deep Network

LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

Contrastive Integrated Gradients: A Feature Attribution-Based Method for Explaining Whole Slide Image Classification

SafeMIL: Learning Offline Safe Imitation Policy from Non-Preferred Trajectories

Dynamic Sparsity: Challenging Common Sparsity Assumptions for Learning World Models in Robotic Reinforcement Learning Benchmarks

Invisible Triggers, Visible Threats! Road-Style Adversarial Creation Attack for Visual 3D Detection in Autonomous Driving

Intelligence per Watt: Measuring Intelligence Efficiency of Local AI

SPUR: A Plug-and-Play Framework for Integrating Spatial Audio Understanding and Reasoning into Large Audio-Language Models

Exploiting Inter-Session Information with Frequency-enhanced Dual-Path Networks for Sequential Recommendation

EMOD: A Unified EEG Emotion Representation Framework Leveraging V-A Guided Contrastive Learning

AI Assisted AR Assembly: Object Recognition and Computer Vision for Augmented Reality Assisted Assembly

Adaptive Detection of Software Aging under Workload Shift

Pelican-VL 1.0: A Foundation Brain Model for Embodied Intelligence

A filtering scheme for confocal laser endomicroscopy (CLE)-video sequences for self-supervised learning

Reflections on the Reproducibility of Commercial LLM Performance in Empirical Software Engineering Studies

Advantage Shaping as Surrogate Reward Maximization: Unifying Pass@K Policy Gradients

ICL-Router: In-Context Learned Model Representations for LLM Routing

Causal Digital Twins for Cyber-Physical Security: A Framework for Robust Anomaly Detection in Industrial Control Systems

Synthetic Object Compositions for Scalable and Accurate Learning in Detection, Segmentation, and Grounding

An Adaptive Multi Agent Bitcoin Trading System

BanglaTalk: Towards Real-Time Speech Assistance for Bengali Regional Dialects

SecInfer: Preventing Prompt Injection via Inference-time Scaling

Re-FRAME the Meeting Summarization SCOPE: Fact-Based Summarization and Personalization via Questions

Threat Modeling for Enhancing Security of IoT Audio Classification Devices under a Secure Protocols Framework

Efficient Learning-Based Control of a Legged Robot in Lunar Gravity

Adaptive Pareto-Optimal Token Merging for Edge Transformer Models in Semantic Communication

Advanced Torrential Loss Function for Precipitation Forecasting

TTF-VLA: Temporal Token Fusion via Pixel-Attention Integration for Vision-Language-Action Models

$\mathrm{TIME}[t]\subseteq \mathrm{SPACE}[O(\sqrt{t})]$ via Tree Height Compression

Visual Document Understanding and Reasoning: A Multi-Agent Collaboration Framework with Agent-Wise Adaptive Test-Time Scaling

FastDriveVLA: Efficient End-to-End Driving via Plug-and-Play Reconstruction-based Token Pruning

First-Order Error Matters: Accurate Compensation for Quantized Large Language Models

StreamDiT: Real-Time Streaming Text-to-Video Generation

NOCTIS: Novel Object Cyclic Threshold based Instance Segmentation

On the Necessity of Output Distribution Reweighting for Effective Class Unlearning

Orthogonal Soft Pruning for Efficient Class Unlearning

Optimization-Induced Dynamics of Lipschitz Continuity in Neural Networks

Self-supervised Learning of Echocardiographic Video Representations via Online Cluster Distillation

Symmetrical Flow Matching: Unified Image Generation, Segmentation, and Classification with Score-Based Generative Models

Zero-Shot Temporal Interaction Localization for Egocentric Videos

Latent Principle Discovery for Language Model Self-Improvement

Transformer Copilot: Learning from The Mistake Log in LLM Fine-tuning

Towards Effective Federated Graph Foundation Model via Mitigating Knowledge Entanglement

$\textit{New News}$: System-2 Fine-tuning for Robust Integration of New Knowledge

Representation Meets Optimization: Training PINNs and PIKANs for Gray-Box Discovery in Systems Pharmacology

Walk Before You Dance: High-fidelity and Editable Dance Synthesis via Generative Masked Motion Prior

Mining--Gym: A Configurable RL Benchmarking Environment for Truck Dispatch Scheduling

The Empty Chair: Using LLMs to Raise Missing Perspectives in Policy Deliberations

Concept-as-Tree: A Controllable Synthetic Data Framework Makes Stronger Personalized VLMs

FedALT: Federated Fine-Tuning through Adaptive Local Training with Rest-of-World LoRA

RAG-Enhanced Collaborative LLM Agents for Drug Discovery

Text-to-SQL Domain Adaptation via Human-LLM Collaborative Data Annotation

FQ-PETR: Fully Quantized Position Embedding Transformation for Multi-View 3D Object Detection

Enhanced Structured Lasso Pruning with Class-wise Information

Efficient Image Restoration via Latent Consistency Flow Matching

Adapt-Pruner: Adaptive Structural Pruning for Efficient Small Language Model Training

The Temporal Trap: Entanglement in Pre-Trained Visual Representations for Visuomotor Policy Learning

LDC: Learning to Generate Research Idea with Dynamic Control

MOSABench: Multi-Object Sentiment Analysis Benchmark for Evaluating Multimodal Large Language Models Understanding of Complex Image

DreamRunner: Fine-Grained Compositional Story-to-Video Generation with Retrieval-Augmented Motion Adaptation

Strada-LLM: Graph LLM for traffic prediction

Towards Formalizing Spuriousness of Biased Datasets Using Partial Information Decomposition

Are language models rational? The case of coherence norms and belief revision

Posterior Label Smoothing for Node Classification

Partial Information Decomposition for Data Interpretability and Feature Selection

Fractured Glass, Failing Cameras: Simulating Physics-Based Adversarial Samples for Autonomous Driving Systems

MoPE: Mixture of Prompt Experts for Parameter-Efficient and Scalable Multimodal Fusion

Survey in Characterization of Semantic Change

GreatSplicing: A Semantically Rich Splicing Dataset

Bi-Level Contextual Bandits for Individualized Resource Allocation under Delayed Feedback

Strategic Opponent Modeling with Graph Neural Networks, Deep Reinforcement Learning and Probabilistic Topic Modeling

Intelligence Foundation Model: A New Perspective to Approach Artificial General Intelligence

Enhancing the Medical Context-Awareness Ability of LLMs via Multifaceted Self-Refinement Learning

OIDA-QA: A Multimodal Benchmark for Analyzing the Opioid Industry Documents Archive

Proceedings of the Second International Workshop on Next-Generation Language Models for Knowledge Representation and Reasoning (NeLaMKRR 2025)

An Efficient Training Pipeline for Reasoning Graphical User Interface Agents

Thinker: Training LLMs in Hierarchical Thinking for Deep Search via Multi-Turn Interaction

Data Complexity of Querying Description Logic Knowledge Bases under Cost-Based Semantics

Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectives

Synthetic Data-Driven Prompt Tuning for Financial QA over Tables and Documents

Large Language Model-assisted Autonomous Vehicle Recovery from Immobilization

VoiceAgentEval: A Dual-Dimensional Benchmark for Expert-Level Intelligent Voice-Agent Evaluation of Xbench's Professional-Aligned Series

Practical, Utilitarian Algorithm Configuration

Clutch Control: An Attention-based Combinatorial Bandit for Efficient Mutation in JavaScript Engine Fuzzing

The Carbon Footprint Wizard: A Knowledge-Augmented AI Interface for Streamlining Food Carbon Footprint Analysis

NetGent: Agent-Based Automation of Network Application Workflows

CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge

Optimizing Multi-Tier Supply Chain Ordering with LNN+XGBoost: Mitigating the Bullwhip Effect

Efficient Story Point Estimation With Comparative Learning

Comprehension Without Competence: Architectural Limits of LLMs in Symbolic Computation and Reasoning

MUDAS: Mote-scale Unsupervised Domain Adaptation in Multi-label Sound Classification

Modeling the Diachronic Evolution of Legal Norms: An LRMoo-Based, Component-Level, Event-Centric Approach to Legal Knowledge Graphs

Sensory-Motor Control with Large Language Models via Iterative Policy Refinement

Do LLMs trust AI regulation? Emerging behaviour of game-theoretic LLM agents

Created by

Haebom

저자

Alessio Buscemi, Daniele Proverbio, Paolo Bova, Nataliya Balabanova, Adeela Bashir, Theodor Cimpeanu, Henrique Correia da Fonseca, Manh Hong Duong, Elias Fernandez Domingos, Antonio M. Fernandes, Marcus Krellner, Ndidi Bianca Ogbo, Simon T. Powers, Fernando P. Santos, Zia Ush Shamszaman, Zhao Song, Alessandro Di Stefano, The Anh Han

개요

본 논문은 진화 게임 이론(EGT) 프레임워크 내에 대규모 언어 모델(LLM) 에이전트를 통합하여 AI 개발 생태계 내 신뢰 및 협력 증진의 중요성을 조사합니다. AI 개발자, 규제 기관, 사용자 간의 전략적 선택을 다양한 규제 시나리오 하에서 모델링하여, 각 행위자가 직면하는 딜레마를 정량적으로 모델링합니다. LLM은 게임의 복잡성과 뉘앙스를 더하고 반복 게임 및 성격 특성 통합을 가능하게 합니다. 연구 결과, 전략적 AI 에이전트는 순수 게임 이론 에이전트보다 더 "비관적"(불신하고 결함이 있는) 태도를 취하는 경향이 있음을 보여줍니다. 사용자의 완전한 신뢰의 경우, 인센티브는 효과적인 규제를 촉진하는 데 효과적이지만, 조건부 신뢰는 "사회적 계약"을 악화시킬 수 있습니다. 따라서 사용자의 신뢰와 규제 기관의 평판 사이의 선순환적인 피드백을 구축하는 것이 개발자들이 안전한 AI를 만드는 데 중요한 것으로 나타났습니다. 그러나 이러한 신뢰가 나타나는 수준은 테스트에 사용되는 특정 LLM에 따라 달라질 수 있습니다. 본 연구 결과는 AI 규제 시스템에 대한 지침을 제공하고, 규제 자체를 지원하는 데 LLM 에이전트가 사용되는 경우 전략적 LLM 에이전트의 결과를 예측하는 데 도움이 됩니다.

시사점, 한계점

•

시사점:

◦

진화 게임 이론과 LLM을 결합하여 AI 개발 생태계 내 신뢰 및 협력에 대한 통찰력을 제공합니다.

◦

사용자의 신뢰와 규제 기관의 평판 간의 선순환적 피드백의 중요성을 강조합니다.

◦

AI 규제 시스템 설계 및 전략적 LLM 에이전트의 결과 예측에 대한 지침을 제공합니다.

◦

전략적 AI 에이전트의 비관적인 경향을 밝혀냄으로써 AI 개발 및 규제에 대한 새로운 이해를 제공합니다.

•

한계점:

◦

사용된 특정 LLM에 따라 결과가 달라질 수 있습니다.

◦

모델의 복잡성으로 인해 결과 해석에 어려움이 있을 수 있습니다.

◦

실제 세계의 복잡성을 완전히 반영하지 못할 수 있습니다.

◦

특정 LLM에 대한 의존성으로 일반화 가능성에 제한이 있을 수 있습니다.

Made with Slashpage