[공지사항]을 빙자한 안부와 근황

Show more

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.

SystolicAttention: Fusing FlashAttention within a Single Systolic Array

Automated Novelty Evaluation of Academic Paper: A Collaborative Approach Integrating Human and Large Language Model Knowledge

"Is it always watching? Is it always listening?" Exploring Contextual Privacy and Security Concerns Toward Domestic Social Robots

A Group Theoretic Analysis of the Symmetries Underlying Base Addition and Their Learnability by Neural Networks

GHPO: Adaptive Guidance for Stable and Efficient LLM Reinforcement Learning

Extension OL-MDISF: Online Learning from Mix-Typed, Drifted, and Incomplete Streaming Features

When and Where do Data Poisons Attack Textual Inversion?

Truth Sleuth and Trend Bender: AI Agents to fact-check YouTube videos and influence opinions

NLP Meets the World: Toward Improving Conversations With the Public About Natural Language Processing Research

Accurate generation of chemical reaction transition states by conditional flow matching

Benchmarking and Evaluation of AI Models in Biology: Outcomes and Recommendations from the CZI Virtual Cells Workshop

A PBN-RL-XAI Framework for Discovering a "Hit-and-Run" Therapeutic Strategy in Melanoma

NeuTSFlow: Modeling Continuous Functions Behind Time Series Forecasting

THOR: Transformer Heuristics for On-Demand Retrieval

Towards Agentic RAG with Deep Reasoning: A Survey of RAG-Reasoning Systems in LLMs

Bridging Literature and the Universe Via A Multi-Agent Large Language Model System

Magneto-radiative modelling and artificial neural network optimization of biofluid flow in a stenosed arterial domain

Symbiosis: Multi-Adapter Inference and Fine-Tuning

Rethinking Data Protection in the (Generative) Artificial Intelligence Era

SoK: Semantic Privacy in Large Language Models

FedRef: Communication-Efficient Bayesian Fine Tuning with Reference Model

Predictable Scale: Part II, Farseer: A Refined Scaling Law in Large Language Models

Position Prediction Self-Supervised Learning for Multimodal Satellite Imagery Semantic Segmentation

ScaleRTL: Scaling LLMs with Reasoning Data and Test-Time Compute for Accurate RTL Code Generation

HueManity: Probing Fine-Grained Visual Perception in MLLMs

AKReF: An argumentative knowledge representation framework for structured argumentation

Large Language Models Often Know When They Are Being Evaluated

Dynamic Risk Assessments for Offensive Cybersecurity Agents

How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference

Diffused Responsibility: Analyzing the Energy Consumption of Generative Text-to-Audio Diffusion Models

Flow-GRPO: Training Flow Matching Models via Online RL

On the Need for a Statistical Foundation in Scenario-Based Testing of Autonomous Vehicles

What's Pulling the Strings? Evaluating Integrity and Attribution in AI Training and Inference through Concept Shift

TD-EVAL: Revisiting Task-Oriented Dialogue Evaluation by Combining Turn-Level Precision with Dialogue-Level Comparisons

MobileCity: An Efficient Framework for Large-Scale Urban Behavior Simulation

Semantic Adapter for Universal Text Embeddings: Diagnosing and Mitigating Negation Blindness to Enhance Universality

Leveraging LLMs for User Stories in AI Systems: UStAI Dataset

Large Language Models are Unreliable for Cyber Threat Intelligence

AnnoPage Dataset: Dataset of Non-Textual Elements in Documents with Fine-Grained Categorization

A Thorough Assessment of the Non-IID Data Impact in Federated Learning

Visual Position Prompt for MLLM based Visual Grounding

Neurons: Emulating the Human Visual Cortex Improves Fidelity and Interpretability in fMRI-to-Video Reconstruction

FADE: Why Bad Descriptions Happen to Good Features

FlipConcept: Tuning-Free Multi-Concept Personalization for Text-to-Image Generation

LUMINA-Net: Low-light Upgrade through Multi-stage Illumination and Noise Adaptation Network for Image Enhancement

Towards Geo-Culturally Grounded LLM Generations

Learning to Reason at the Frontier of Learnability

Flexible and Efficient Grammar-Constrained Decoding

PATCH: a deep learning method to assess heterogeneity of artistic practice in historical paintings

The Impact of Modern AI in Metadata Management

Learning an Effective Premise Retrieval Model for Efficient Mathematical Formalization

ChipAlign: Instruction Alignment in Large Language Models for Chip Design via Geodesic Interpolation

Many Objective Problems Where Crossover is Provably Essential

Patherea: Cell Detection and Classification for the 2020s

ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy

TextDestroyer: A Training- and Annotation-Free Diffusion Method for Destroying Anomal Text from Images

Quantifying calibration error in modern neural networks through evidence based theory

Multi-view biomedical foundation models for molecule-target and property prediction

Reinforced Imitative Trajectory Planning for Urban Automated Driving

Distilling Invariant Representations with Dual Augmentation

Tackling the Abstraction and Reasoning Corpus with Vision Transformers: the Importance of 2D Representation, Positions, and Objects

Linearly-Interpretable Concept Embedding Models for Text Analysis

Towards Understanding Link Predictor Generalizability Under Distribution Shifts

StreakNet-Arch: An Anti-scattering Network-based Architecture for Underwater Carrier LiDAR-Radar Imaging

Enhancing Trust in Autonomous Agents: An Architecture for Accountability and Explainability through Blockchain and Large Language Models

On the Statistical Properties of Generative Adversarial Models for Low Intrinsic Data Dimension

Programming Distributed Collective Processes in the eXchange Calculus

Holistic analysis on the sustainability of Federated Learning across AI product lifecycle

Mathematical Introduction to Deep Learning: Methods, Implementations, and Theory

Epic-Sounds: A Large-scale Dataset of Actions That Sound

From Semantic Web and MAS to Agentic AI: A Unified Narrative of the Web of Agents

On Gradual Semantics for Assumption-Based Argumentation

The Challenge of Teaching Reasoning to LLMs Without RL or Distillation

Continuous Classification Aggregation

Can Prompt Difficulty be Online Predicted for Accelerating RL Finetuning of Reasoning Models?

macOSWorld: A Multilingual Interactive Benchmark for GUI Agents

GeoChain: Multimodal Chain-of-Thought for Geographic Reasoning

Lost in Transmission: When and Why LLMs Fail to Reason Globally

A Survey of Frontiers in LLM Reasoning: Inference Scaling, Learning to Reason, and Agentic Systems

System 0/1/2/3: Quad-process theory for multi-timescale embodied collective cognitive systems

Practical Principles for AI Cost and Compute Accounting

Generative Emergent Communication: Large Language Model is a Collective World Model

Proactive Agents for Multi-Turn Text-to-Image Generation Under Uncertainty

Learning Lifted STRIPS Models from Action Traces Alone: A Simple, General, and Scalable Solution

Hallucination Detox: Sensitivity Dropout (SenD) for Large Language Model Training

Life, uh, Finds a Way: Hyperadaptability by Behavioral Search

Governance of Generative Artificial Intelligence for Companies

RACER: Rational Artificial Intelligence Car-following-model Enhanced by Reality

Artificial Intelligence Governance for Businesses

Interpreting Radiologist's Intention from Eye Movements in Chest X-ray Diagnosis

S2WTM: Spherical Sliced-Wasserstein Autoencoder for Topic Modeling

LLM-Based Config Synthesis requires Disambiguation

Characterizing State Space Model (SSM) and SSM-Transformer Hybrid Language Model Performance with Long Context Length

EgoVLA: Learning Vision-Language-Action Models from Egocentric Human Videos

Can We Predict Alignment Before Models Finish Thinking? Towards Monitoring Misaligned Reasoning Models

Unit-Based Histopathology Tissue Segmentation via Multi-Level Feature Representation

Advancing Retrieval-Augmented Generation for Structured Enterprise and Internal Data

Mixture of Raytraced Experts

QuRe: Query-Relevant Retrieval through Hard Negative Sampling in Composed Image Retrieval

AutoVDC: Automated Vision Data Cleaning Using Vision-Language Models

TRAIL: Trace Reasoning and Agentic Issue Localization

Created by

Haebom

저자

Darshan Deshpande, Varun Gangal, Hersh Mehta, Jitin Krishnan, Anand Kannappan, Rebecca Qian

개요

본 논문은 다양한 분야에서 에이전트 워크플로우의 채택 증가에 따라 이러한 시스템이 생성하는 복잡한 추적을 확장 가능하고 체계적으로 평가할 필요성을 강조합니다. 기존의 평가 방법은 긴 워크플로우 추적에 대한 수동적이고 도메인 특정적인 인간 분석에 의존하는데, 이는 에이전트 출력의 복잡성과 양이 증가함에 따라 확장성이 부족합니다. 본 논문에서는 에이전트 워크플로우 추적을 위한 강력하고 동적인 평가 방법의 필요성을 명확히 하고, 에이전트 시스템에서 발생하는 오류 유형에 대한 공식적인 분류 체계를 제시하며, 이 분류 체계를 기반으로 구축된 148개의 대규모 인간 주석 추적(TRAIL) 데이터셋을 소개합니다. TRAIL 데이터셋은 소프트웨어 엔지니어링 및 개방형 세계 정보 검색과 같은 실제 응용 프로그램에 중점을 두고 단일 및 다중 에이전트 시스템 모두에서 추적을 큐레이팅하여 생태학적 타당성을 보장합니다. 평가 결과, 최신 장문 컨텍스트 LLMs는 추적 디버깅에서 성능이 저조하며, 최고 성능 모델인 Gemini-2.5-pro조차 TRAIL에서 11%의 정확도만 기록했습니다. 본 논문에서는 TRAIL 데이터셋과 코드를 공개하여 에이전트 워크플로우에 대한 확장 가능한 평가 분야의 미래 연구를 지원하고 가속화합니다.

시사점, 한계점

•

시사점: 에이전트 워크플로우의 확장 가능하고 체계적인 평가를 위한 새로운 방법론과 데이터셋(TRAIL)을 제시. 에이전트 시스템 오류 분석을 위한 공식적인 오류 유형 분류 체계 제공. 현존하는 대규모 언어 모델의 에이전트 워크플로우 추적 디버깅 성능의 한계를 밝힘. 공개된 데이터셋과 코드를 통해 향후 연구를 가속화.

•

한계점: 현재 TRAIL 데이터셋의 규모(148개 추적)가 향후 더욱 복잡하고 다양한 에이전트 시스템 평가에 충분한지에 대한 검토 필요. 평가에 사용된 모델의 종류가 제한적이므로, 더욱 다양한 모델에 대한 평가가 필요. 인간 주석의 주관성으로 인한 오류 가능성 존재. 실제 세계의 다양한 에이전트 시스템을 완벽히 반영하지 못할 가능성.

Made with Slashpage