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.

Probing Experts' Perspectives on AI-Assisted Public Speaking Training

Learning Pole Structures of Hadronic States using Predictive Uncertainty Estimation

Neural Concept Verifier: Scaling Prover-Verifier Games via Concept Encodings

Hallucination Stations: On Some Basic Limitations of Transformer-Based Language Models

Objectomaly: Objectness-Aware Refinement for OoD Segmentation with Structural Consistency and Boundary Precision

KeyRe-ID: Keypoint-Guided Person Re-Identification using Part-Aware Representation in Videos

Squeeze the Soaked Sponge: Efficient Off-policy Reinforcement Finetuning for Large Language Model

The Dark Side of LLMs Agent-based Attacks for Complete Computer Takeover

Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

Hyperspectral Anomaly Detection Methods: A Survey and Comparative Study

EmissionNet: Air Quality Pollution Forecasting for Agriculture

The role of gain neuromodulation in layer-5 pyramidal neurons

USAD: End-to-End Human Activity Recognition via Diffusion Model with Spatiotemporal Attention

Hita: Holistic Tokenizer for Autoregressive Image Generation

Distributional Soft Actor-Critic with Diffusion Policy

Lighting the Night with Generative Artificial Intelligence

On the Necessity of Output Distribution Reweighting for Effective Class Unlearning

Upgrade or Switch: Do We Need a Next-Gen Trusted Architecture for the Internet of AI Agents?

Language-Grounded Hierarchical Planning and Execution with Multi-Robot 3D Scene Graphs

Grokking Beyond the Euclidean Norm of Model Parameters

An Outlook on the Opportunities and Challenges of Multi-Agent AI Systems

Deep Learning-Based Forecasting of Boarding Patient Counts to Address ED Overcrowding

An Exploration of Default Images in Text-to-Image Generation

Automatic Curriculum Learning for Driving Scenarios: Towards Robust and Efficient Reinforcement Learning

TPK: Trustworthy Trajectory Prediction Integrating Prior Knowledge For Interpretability and Kinematic Feasibility

Boundary-Guided Trajectory Prediction for Road Aware and Physically Feasible Autonomous Driving

Balancing Progress and Safety: A Novel Risk-Aware Objective for RL in Autonomous Driving

TS-SNN: Temporal Shift Module for Spiking Neural Networks

Red Teaming Large Language Models for Healthcare

One-Pass to Reason: Token Duplication and Block-Sparse Mask for Efficient Fine-Tuning on Multi-Turn Reasoning

AI Safety Should Prioritize the Future of Work

MGT: Extending Virtual Try-Off to Multi-Garment Scenarios

MedSegFactory: Text-Guided Generation of Medical Image-Mask Pairs

Generative Retrieval and Alignment Model: A New Paradigm for E-commerce Retrieval

Using AI to Summarize US Presidential Campaign TV Advertisement Videos, 1952-2012

REGEN: A Dataset and Benchmarks with Natural Language Critiques and Narratives

GTR: Guided Thought Reinforcement Prevents Thought Collapse in RL-based VLM Agent Training

Field Matching: an Electrostatic Paradigm to Generate and Transfer Data

SP$^2$T: Sparse Proxy Attention for Dual-stream Point Transformer

PIAD-SRNN: Physics-Informed Adaptive Decomposition in State-Space RNN

FonTS: Text Rendering with Typography and Style Controls

On the Principles of ReLU Networks with One Hidden Layer

Compositional Risk Minimization

End-to-end multi-channel speaker extraction and binaural speech synthesis

Post-hoc Study of Climate Microtargeting on Social Media Ads with LLMs: Thematic Insights and Fairness Evaluation

Downscaling Extreme Precipitation with Wasserstein Regularized Diffusion

Quantifying Context Bias in Domain Adaptation for Object Detection

An Empirical Study of Validating Synthetic Data for Formula Generation

Evaluating Implicit Bias in Large Language Models by Attacking From a Psychometric Perspective

HeSum: a Novel Dataset for Abstractive Text Summarization in Hebrew

SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths

GoalNet: Goal Areas Oriented Pedestrian Trajectory Prediction

Large Language Models in Mental Health Care: a Scoping Review

SDR-GAIN: A High Real-Time Occluded Pedestrian Pose Completion Method for Autonomous Driving

Temporal Motifs for Financial Networks: A Study on Mercari, JPMC, and Venmo Platforms

Measuring AI Alignment with Human Flourishing

StarDojo: Benchmarking Open-Ended Behaviors of Agentic Multimodal LLMs in Production-Living Simulations with Stardew Valley

Open Source Planning & Control System with Language Agents for Autonomous Scientific Discovery

Discovering Algorithms with Computational Language Processing

A Hybrid SMT-NRA Solver: Integrating 2D Cell-Jump-Based Local Search, MCSAT and OpenCAD

A taxonomy of epistemic injustice in the context of AI and the case for generative hermeneutical erasure

Bandit-Based Prompt Design Strategy Selection Improves Prompt Optimizers

Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces

AI Delegates with a Dual Focus: Ensuring Privacy and Strategic Self-Disclosure

Text2BIM: Generating Building Models Using a Large Language Model-based Multi-Agent Framework

Interpreting systems as solving POMDPs: a step towards a formal understanding of agency

Lumos-1: On Autoregressive Video Generation from a Unified Model Perspective

NeuralOS: Towards Simulating Operating Systems via Neural Generative Models

KV Cache Steering for Inducing Reasoning in Small Language Models

Optimistic Exploration for Risk-Averse Constrained Reinforcement Learning

On Barriers to Archival Audio Processing

A Hybrid Multi-Well Hopfield-CNN with Feature Extraction and K-Means for MNIST Classification

Compress Any Segment Anything Model (SAM)

Penalizing Infeasible Actions and Reward Scaling in Reinforcement Learning with Offline Data

Geo-ORBIT: A Federated Digital Twin Framework for Scene-Adaptive Lane Geometry Detection

Adaptive Nonlinear Vector Autoregression: Robust Forecasting for Noisy Chaotic Time Series

Catastrophic Forgetting Mitigation Through Plateau Phase Activity Profiling

Dually Hierarchical Drift Adaptation for Online Configuration Performance Learning

Monitoring Risks in Test-Time Adaptation

Multilingual Multimodal Software Developer for Code Generation

KG-Attention: Knowledge Graph-Guided Attention at Test-Time via Bidirectional Information Aggregation

ONION: A Multi-Layered Framework for Participatory ER Design

A Personalised Formal Verification Framework for Monitoring Activities of Daily Living of Older Adults Living Independently in Their Homes

MoSAiC: Multi-Modal Multi-Label Supervision-Aware Contrastive Learning for Remote Sensing

KELPS: A Framework for Verified Multi-Language Autoformalization via Semantic-Syntactic Alignment

Safe Deep Reinforcement Learning for Resource Allocation with Peak Age of Information Violation Guarantees

DatasetAgent: A Novel Multi-Agent System for Auto-Constructing Datasets from Real-World Images

Scaling Attention to Very Long Sequences in Linear Time with Wavelet-Enhanced Random Spectral Attention (WERSA)

Normalized vs Diplomatic Annotation: A Case Study of Automatic Information Extraction from Handwritten Uruguayan Birth Certificates

Adaptive Framework for Ambient Intelligence in Rehabilitation Assistance

A comprehensive study of LLM-based argument classification: from LLAMA through GPT-4o to Deepseek-R1

Towards Collaborative Fairness in Federated Learning Under Imbalanced Covariate Shift

Generating Proto-Personas through Prompt Engineering: A Case Study on Efficiency, Effectiveness and Empathy

To Trade or Not to Trade: An Agentic Approach to Estimating Market Risk Improves Trading Decisions

A Multi-Modal Fusion Framework for Brain Tumor Segmentation Based on 3D Spatial-Language-Vision Integration and Bidirectional Interactive Attention Mechanism

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

RadiomicsRetrieval: A Customizable Framework for Medical Image Retrieval Using Radiomics Features

White-Basilisk: A Hybrid Model for Code Vulnerability Detection

MIDI-VALLE: Improving Expressive Piano Performance Synthesis Through Neural Codec Language Modelling

PromotionGo at SemEval-2025 Task 11: A Feature-Centric Framework for Cross-Lingual Multi-Emotion Detection in Short Texts

ST-WebAgentBench: A Benchmark for Evaluating Safety and Trustworthiness in Web Agents

Created by

Haebom

저자

Ido Levy, Ben Wiesel, Sami Marreed, Alon Oved, Avi Yaeli, Segev Shlomov

개요

본 논문은 기존 웹 에이전트 평가 벤치마크가 작업 완료 여부만 평가하고 안전성과 신뢰성을 고려하지 않는다는 점을 지적하며, 기업 환경에서의 안전하고 신뢰할 수 있는 웹 에이전트 배포를 위한 새로운 벤치마크인 ST-WebAgentBench를 제시합니다. ST-WebAgentBench는 222개의 현실적인 기업 시나리오 기반 작업과 각 작업에 대한 안전 및 신뢰성(ST) 정책, 그리고 사용자 동의, 강건성 등 6가지 차원에 걸친 평가 점수를 제공합니다. 기존의 작업 완료율 외에, 모든 정책을 준수하는 완료만을 인정하는 Completion Under Policy (CuP) 지표와 ST 위반을 정량화하는 Risk Ratio 지표를 제안하며, 세 개의 최신 웹 에이전트를 평가하여 평균 CuP가 명목상 완료율의 3분의 2 미만임을 밝히고, 중대한 안전성 문제를 드러냅니다. 코드, 평가 템플릿, 정책 작성 인터페이스를 공개하여 실용적인 웹 에이전트 배포를 위한 첫걸음을 제공합니다.

시사점, 한계점

•

시사점:

◦

기존 웹 에이전트 평가의 한계를 극복하고 안전성 및 신뢰성을 고려한 새로운 평가 기준 제시.

◦

현실적인 기업 시나리오를 반영한 222개의 다양한 작업 제공.

◦

CuP 및 Risk Ratio 지표를 통해 안전성 및 신뢰성 위험을 정량적으로 평가 가능.

◦

오픈소스로 공개되어 웹 에이전트 개발 및 평가에 대한 접근성 향상.

◦

최신 웹 에이전트의 안전성 문제를 실증적으로 밝힘으로써 향후 연구 방향 제시.

•

한계점:

◦

제시된 222개의 작업이 모든 기업 환경을 완벽히 포괄하는지는 추가적인 검증 필요.

◦

ST 정책의 정의 및 적용 방식에 대한 주관성 존재 가능성.

◦

평가 대상 에이전트의 종류 및 수가 제한적일 수 있음.

◦

장기적인 안전성 및 신뢰성 평가에 대한 고려 부족.

Made with Slashpage