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.

AutoChemSchematic AI: A Closed-Loop, Physics-Aware Agentic Framework for Auto-Generating Chemical Process and Instrumentation Diagrams

LlamaRL: A Distributed Asynchronous Reinforcement Learning Framework for Efficient Large-scale LLM Training

DLP: Dynamic Layerwise Pruning in Large Language Models

MedHELM: Holistic Evaluation of Large Language Models for Medical Tasks

Estimating LLM Consistency: A User Baseline vs Surrogate Metrics

Mind the Gap: A Practical Attack on GGUF Quantization

Bounded Rationality for LLMs: Satisficing Alignment at Inference-Time

GSO: Challenging Software Optimization Tasks for Evaluating SWE-Agents

Active Layer-Contrastive Decoding Reduces Hallucination in Large Language Model Generation

Jigsaw-R1: A Study of Rule-based Visual Reinforcement Learning with Jigsaw Puzzles

Cognitive Guardrails for Open-World Decision Making in Autonomous Drone Swarms

SWE-bench Goes Live!

Afterburner: Reinforcement Learning Facilitates Self-Improving Code Efficiency Optimization

Matryoshka Model Learning for Improved Elastic Student Models

Context-Robust Knowledge Editing for Language Models

HiLDe: Intentional Code Generation via Human-in-the-Loop Decoding

FAMA: The First Large-Scale Open-Science Speech Foundation Model for English and Italian

FastTD3: Simple, Fast, and Capable Reinforcement Learning for Humanoid Control

Judging Quality Across Languages: A Multilingual Approach to Pretraining Data Filtering with Language Models

iDSE: Navigating Design Space Exploration in High-Level Synthesis Using LLMs

Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection

RePaViT: Scalable Vision Transformer Acceleration via Structural Reparameterization on Feedforward Network Layers

More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models

Hume: Introducing System-2 Thinking in Visual-Language-Action Model

Universal Value-Function Uncertainties

How Do Transformers Learn Variable Binding in Symbolic Programs?

Adversarial bandit optimization for approximately linear functions

Hierarchical Retrieval with Evidence Curation for Open-Domain Financial Question Answering on Standardized Documents

Homophily Enhanced Graph Domain Adaptation

NeuSym-RAG: Hybrid Neural Symbolic Retrieval with Multiview Structuring for PDF Question Answering

SoloSpeech: Enhancing Intelligibility and Quality in Target Speech Extraction through a Cascaded Generative Pipeline

An Interpretable Representation Learning Approach for Diffusion Tensor Imaging

InfoChartQA: A Benchmark for Multimodal Question Answering on Infographic Charts

Moderating Harm: Benchmarking Large Language Models for Cyberbullying Detection in YouTube Comments

Security Concerns for Large Language Models: A Survey

A Survey of LLM $\times$ DATA

Task-Optimized Convolutional Recurrent Networks Align with Tactile Processing in the Rodent Brain

NMCSE: Noise-Robust Multi-Modal Coupling Signal Estimation Method via Optimal Transport for Cardiovascular Disease Detection

FRIREN: Beyond Trajectories -- A Spectral Lens on Time

A Fully Generative Motivational Interviewing Counsellor Chatbot for Moving Smokers Towards the Decision to Quit

TrimR: Verifier-based Training-Free Thinking Compression for Efficient Test-Time Scaling

Beyond Face Swapping: A Diffusion-Based Digital Human Benchmark for Multimodal Deepfake Detection

Attributing Response to Context: A Jensen-Shannon Divergence Driven Mechanistic Study of Context Attribution in Retrieval-Augmented Generation

QuickVideo: Real-Time Long Video Understanding with System Algorithm Co-Design

Replay Attacks Against Audio Deepfake Detection

Forensic deepfake audio detection using segmental speech features

A3 : an Analytical Low-Rank Approximation Framework for Attention

A Survey of 3D Reconstruction with Event Cameras

Confabulation dynamics in a reservoir computer: Filling in the gaps with untrained attractors

WATCH: Adaptive Monitoring for AI Deployments via Weighted-Conformal Martingales

Deep Learning Framework for Infrastructure Maintenance: Crack Detection and High-Resolution Imaging of Infrastructure Surfaces

GAME: Learning Multimodal Interactions via Graph Structures for Personality Trait Estimation

LENSLLM: Unveiling Fine-Tuning Dynamics for LLM Selection

Motion-compensated cardiac MRI using low-rank diffeomorphic flow (DMoCo)

OODTE: A Differential Testing Engine for the ONNX Optimizer

Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization

Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video

Depth-Constrained ASV Navigation with Deep RL and Limited Sensing

(Im)possibility of Automated Hallucination Detection in Large Language Models

Unveiling the Lack of LVLM Robustness to Fundamental Visual Variations: Why and Path Forward

Position: An Empirically Grounded Identifiability Theory Will Accelerate Self-Supervised Learning Research

The Hitchhiker's Guide to Program Analysis, Part II: Deep Thoughts by LLMs

The Structural Safety Generalization Problem

Parameterized Synthetic Text Generation with SimpleStories

SD$^2$: Self-Distilled Sparse Drafters

Persona Dynamics: Unveiling the Impact of Personality Traits on Agents in Text-Based Games

Semantic-guided Representation Learning for Multi-Label Recognition

SynWorld: Virtual Scenario Synthesis for Agentic Action Knowledge Refinement

Bridging the Linguistic Divide: A Survey on Leveraging Large Language Models for Machine Translation

A Conformal Risk Control Framework for Granular Word Assessment and Uncertainty Calibration of CLIPScore Quality Estimates

Understanding Inequality of LLM Fact-Checking over Geographic Regions with Agent and Retrieval models

ADS-Edit: A Multimodal Knowledge Editing Dataset for Autonomous Driving Systems

A Survey on Event-driven 3D Reconstruction: Development under Different Categories

VecTrans: Enhancing Compiler Auto-Vectorization through LLM-Assisted Code Transformations

REALM: A Dataset of Real-World LLM Use Cases

Opportunities and Challenges of Frontier Data Governance With Synthetic Data

ARFlow: Human Action-Reaction Flow Matching with Physical Guidance

Position: Beyond Assistance - Reimagining LLMs as Ethical and Adaptive Co-Creators in Mental Health Care

Redefining Toxicity: An Objective and Context-Aware Approach for Stress-Level-Based Detection

A Dual-Directional Context-Aware Test-Time Learning for Text Classification

MentalChat16K: A Benchmark Dataset for Conversational Mental Health Assistance

ClusComp: A Simple Paradigm for Model Compression and Efficient Finetuning

FMNet: Frequency-Assisted Mamba-Like Linear Attention Network for Camouflaged Object Detection

NFIG: Autoregressive Image Generation with Next-Frequency Prediction

GMLM: Bridging Graph Neural Networks and Language Models for Heterophilic Node Classification

CSTRL: Context-Driven Sequential Transfer Learning for Abstractive Radiology Report Summarization

Wanda++: Pruning Large Language Models via Regional Gradients

HoH: A Dynamic Benchmark for Evaluating the Impact of Outdated Information on Retrieval-Augmented Generation

Optimizing Multi-Hop Document Retrieval Through Intermediate Representations

Causally Reliable Concept Bottleneck Models

SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning

A Comprehensive Survey of Machine Unlearning Techniques for Large Language Models

Marco-o1 v2: Towards Widening The Distillation Bottleneck for Reasoning Models

POPGym Arcade: Parallel Pixelated POMDPs

Semantic Integrity Constraints: Declarative Guardrails for AI-Augmented Data Processing Systems

Quantifying First-Order Markov Violations in Noisy Reinforcement Learning: A Causal Discovery Approach

SPD: Sync-Point Drop for Efficient Tensor Parallelism of Large Language Models

Mixture of Structural-and-Textual Retrieval over Text-rich Graph Knowledge Bases

Faithful Logic Embeddings in HOL -- Deep and Shallow

TestNUC: Enhancing Test-Time Computing Approaches and Scaling through Neighboring Unlabeled Data Consistency

BaxBench: Can LLMs Generate Correct and Secure Backends?

Created by

Haebom

저자

Mark Vero, Niels Mundler, Victor Chibotaru, Veselin Raychev, Maximilian Baader, Nikola Jovanovic, Jingxuan He, Martin Vechev

개요

BaxBench는 백엔드 애플리케이션 생성을 위한 새로운 평가 벤치마크입니다. 392개의 과제로 구성되며, 백엔드에 초점을 맞춘 이유는 실제로 중요하고, 대부분의 현대 웹 및 클라우드 소프트웨어의 핵심 구성 요소를 구축하며, 올바르게 구현하기 어렵고, 보안에 중요하기 때문입니다. BaxBench는 생성된 애플리케이션의 기능을 포괄적인 테스트 사례로 검증하고, 엔드투엔드 악용을 실행하여 보안 노출을 평가합니다. 실험 결과, 현재 LLM의 기능과 보안 모두에 중요한 한계가 있음을 보여줍니다. OpenAI의 o1 모델조차도 코드 정확도에서 62%에 불과했으며, 생성된 정확한 프로그램의 약 절반에서 보안 악용이 성공적으로 실행되었습니다. 인기 없는 백엔드 프레임워크에서는 모델이 정확하고 안전한 애플리케이션을 생성하는 데 더 어려움을 겪었습니다. BaxBench의 발전은 LLM을 사용한 자율적이고 안전한 소프트웨어 개발을 향한 중요한 단계를 의미합니다.

시사점, 한계점

•

시사점: LLM을 이용한 백엔드 애플리케이션 자동 생성의 현황과 한계를 명확히 평가할 수 있는 새로운 벤치마크(BaxBench) 제시. LLM 기반 자율적이고 안전한 소프트웨어 개발의 중요한 발전 방향 제시.

•

한계점: 최고 성능 모델조차도 코드 정확도가 62%에 그침. 생성된 정확한 프로그램의 절반 가량에서 보안 취약점 발견. 인기 없는 백엔드 프레임워크에서 성능 저하. LLM의 기능 및 보안 측면에서 개선이 필요함을 시사.

Made with Slashpage