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.

VisPlay: Self-Evolving Vision-Language Models from Images

CompTrack: Information Bottleneck-Guided Low-Rank Dynamic Token Compression for Point Cloud Tracking

Multimodal Evaluation of Russian-language Architectures

Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models

OEMA: Ontology-Enhanced Multi-Agent Collaboration Framework for Zero-Shot Clinical Named Entity Recognition

Finetuning LLMs for Automatic Form Interaction on Web-Browser in Selenium Testing Framework

Kandinsky 5.0: A Family of Foundation Models for Image and Video Generation

Orion: A Unified Visual Agent for Multimodal Perception, Advanced Visual Reasoning and Execution

Can Artificial Intelligence Accelerate Technological Progress? Researchers' Perspectives on AI in Manufacturing and Materials Science

Node-Level Uncertainty Estimation in LLM-Generated SQL

Auditing Google's AI Overviews and Featured Snippets: A Case Study on Baby Care and Pregnancy

Mesh-based Super-resolution of Detonation Flows with Multiscale Graph Transformers

STAMP: Spatial-Temporal Adapter with Multi-Head Pooling

Do Not Merge My Model! Safeguarding Open-Source LLMs Against Unauthorized Model Merging

Statistically Assuring Safety of Control Systems using Ensembles of Safety Filters and Conformal Prediction

LoRA on the Go: Instance-level Dynamic LoRA Selection and Merging

CaberNet: Causal Representation Learning for Cross-Domain HVAC Energy Prediction

Kaggle Chronicles: 15 Years of Competitions, Community and Data Science Innovation

Injecting Falsehoods: Adversarial Man-in-the-Middle Attacks Undermining Factual Recall in LLMs

TabDistill: Distilling Transformers into Neural Nets for Few-Shot Tabular Classification

Diagnosing Hallucination Risk in AI Surgical Decision-Support: A Sequential Framework for Sequential Validation

GAPO: Robust Advantage Estimation for Real-World Code LLMs

Steering Evaluation-Aware Language Models to Act Like They Are Deployed

Practical and Stealthy Touch-Guided Jailbreak Attacks on Deployed Mobile Vision-Language Agents

HiViS: Hiding Visual Tokens from the Drafter for Speculative Decoding in Vision-Language Models

Decoding Deception: Understanding Automatic Speech Recognition Vulnerabilities in Evasion and Poisoning Attacks

Interpretability as Alignment: Making Internal Understanding a Design Principle

How many patients could we save with LLM priors?

CoBA: Counterbias Text Augmentation for Mitigating Various Spurious Correlations via Semantic Triples

Interpreting the Effects of Quantization on LLMs

From Confidence to Collapse in LLM Factual Robustness

PepThink-R1: LLM for Interpretable Cyclic Peptide Optimization with CoT SFT and Reinforcement Learning

Learning to Detect Unknown Jailbreak Attacks in Large Vision-Language Models

MAQuA: Adaptive Question-Asking for Multidimensional Mental Health Screening using Item Response Theory

LLMDistill4Ads: Using Cross-Encoders to Distill from LLM Signals for Advertiser Keyphrase Recommendations

Efficient Solution and Learning of Robust Factored MDPs

To Trust or Not to Trust: On Calibration in ML-based Resource Allocation for Wireless Networks

Parallelism Meets Adaptiveness: Scalable Documents Understanding in Multi-Agent LLM Systems

HAWAII: Hierarchical Visual Knowledge Transfer for Efficient Vision-Language Models

When concept-based XAI is imprecise: Do people distinguish between generalisations and misrepresentations?

Eliciting Reasoning in Language Models with Cognitive Tools

Beyond Bias Scores: Unmasking Vacuous Neutrality in Small Language Models

From Static to Adaptive Defense: Federated Multi-Agent Deep Reinforcement Learning-Driven Moving Target Defense Against DoS Attacks in UAV Swarm Networks

Policy Search, Retrieval, and Composition via Task Similarity in Collaborative Agentic Systems

Fast-DataShapley: Neural Modeling for Training Data Valuation

An Iterative Question-Guided Framework for Knowledge Base Question Answering

A survey of using EHR as real-world evidence for discovering and validating new drug indications

CSI-Bench: A Large-Scale In-the-Wild Dataset for Multi-task WiFi Sensing

A Distributionally Robust Framework for Nuisance in Causal Effect Estimation

BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms

Vector Quantized-Elites: Unsupervised and Problem-Agnostic Quality-Diversity Optimization

CaKE: Circuit-aware Editing Enables Generalizable Knowledge Learners

A Closer Look at Adversarial Suffix Learning for Jailbreaking LLMs: Augmented Adversarial Trigger Learning

CleverDistiller: Simple and Spatially Consistent Cross-modal Distillation

LLMInit: A Free Lunch from Large Language Models for Selective Initialization of Recommendation

Securing Smart Contract Languages with a Unified Agentic Framework for Vulnerability Repair in Solidity and Move

TRADES: Generating Realistic Market Simulations with Diffusion Models

Recent Advances in Discrete Speech Tokens: A Review

Oracular Programming: A Modular Foundation for Building LLM-Enabled Software

KVTuner: Sensitivity-Aware Layer-Wise Mixed-Precision KV Cache Quantization for Efficient and Nearly Lossless LLM Inference

RAPID: Robust and Agile Planner Using Inverse Reinforcement Learning for Vision-Based Drone Navigation

OmniThink: Expanding Knowledge Boundaries in Machine Writing through Thinking

Efficient Architectures for High Resolution Vision-Language Models

MMVA: Multimodal Matching Based on Valence and Arousal across Images, Music, and Musical Captions

Automatically Detecting Online Deceptive Patterns

LEARNER: Contrastive Pretraining for Learning Fine-Grained Patient Progression from Coarse Inter-Patient Labels

Atomic Calibration of LLMs in Long-Form Generations

TopoTune : A Framework for Generalized Combinatorial Complex Neural Networks

Asymptotic and Finite Sample Analysis of Nonexpansive Stochastic Approximations with Markovian Noise

Introducing DEFORMISE: A deep learning framework for dementia diagnosis in the elderly using optimized MRI slice selection

KWT-Tiny: RISC-V Accelerated, Embedded Keyword Spotting Transformer

Provably Robust Pre-Trained Ensembles for Biomarker-Based Cancer Classification

DiffuSyn Bench: Evaluating Vision-Language Models on Real-World Complexities with Diffusion-Generated Synthetic Benchmarks

Property-guided Inverse Design of Metal-Organic Frameworks Using Quantum Natural Language Processing

Can LLMs Replace Economic Choice Prediction Labs? The Case of Language-based Persuasion Games

LSAP: Rethinking Inversion Fidelity, Perception and Editability in GAN Latent Space

As If We've Met Before: LLMs Exhibit Certainty in Recognizing Seen Files

SafeRBench: A Comprehensive Benchmark for Safety Assessment in Large Reasoning Models

Benchmarking Multi-Step Legal Reasoning and Analyzing Chain-of-Thought Effects in Large Language Models

Multi-dimensional Data Analysis and Applications Basing on LLM Agents and Knowledge Graph Interactions

Towards Efficient Multimodal Unified Reasoning Model via Model Merging

FATHOMS-RAG: A Framework for the Assessment of Thinking and Observation in Multimodal Systems that use Retrieval Augmented Generation

Probing the Critical Point (CritPt) of AI Reasoning: a Frontier Physics Research Benchmark

Evaluating Multimodal Large Language Models with Daily Composite Tasks in Home Environments

Taming Uncertainty via Automation: Observing, Analyzing, and Optimizing Agentic AI Systems

Constraint-Guided Prediction Refinement via Deterministic Diffusion Trajectories

The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity

PoE-World: Compositional World Modeling with Products of Programmatic Experts

Bridging the Gap in XAI-Why Reliable Metrics Matter for Explainability and Compliance

Dataset Distillation for Pre-Trained Self-Supervised Vision Models

Thinking-while-Generating: Interleaving Textual Reasoning throughout Visual Generation

Taming the Long-Tail: Efficient Reasoning RL Training with Adaptive Drafter

Dexterity from Smart Lenses: Multi-Fingered Robot Manipulation with In-the-Wild Human Demonstrations

Teacher-Guided One-Shot Pruning via Context-Aware Knowledge Distillation

Evolution Strategies at the Hyperscale

Faster Certified Symmetry Breaking Using Orders With Auxiliary Variables

Stabilizing Policy Gradient Methods via Reward Profiling

SAM 3D: 3Dfy Anything in Images

Improving Long-Tailed Object Detection with Balanced Group Softmax and Metric Learning

Generative AI for Enhanced Wildfire Detection: Bridging the Synthetic-Real Domain Gap

Practical and Stealthy Touch-Guided Jailbreak Attacks on Deployed Mobile Vision-Language Agents

Created by

Haebom

저자

Renhua Ding, Xiao Yang, Zhengwei Fang, Jun Luo, Kun He, Jun Zhu

개요

본 논문은 대규모 비전-언어 모델(LVLM)을 활용한 자율 모바일 에이전트의 스마트폰 인터페이스 운영에 존재하는 보안 취약점을 연구합니다. 특히, 기존 연구의 한계를 극복하고 실용적이고 은밀한 jailbreak 공격 프레임워크를 제시합니다. 이 프레임워크는 (i) 비권한적 인식 손상, (ii) 에이전트 귀속 활성화, (iii) 효율적인 원샷 jailbreak의 세 가지 핵심 구성 요소를 포함합니다. 또한, Android 애플리케이션 개발 및 프롬프트 주입 데이터셋 구축을 통해 공격의 유효성을 평가했습니다. 다양한 LVLM 백엔드에서 높은 공격 성공률을 확인하여 자율 스마트폰 운영에 대한 근본적인 보안 취약성을 드러냈습니다.

시사점, 한계점

•

시사점:

◦

LVLM 기반 모바일 에이전트의 보안 취약성을 실증적으로 입증하고, 실용적인 공격 프레임워크를 제시함.

◦

은밀성을 높인 공격 기법을 통해 실제 환경에서의 위협 가능성을 제기함.

◦

다양한 LVLM 백엔드에서 공격 성공을 확인하여 광범위한 취약성을 시사함.

◦

자율 스마트폰 운영의 보안에 대한 중요한 시사점을 제시하고, 관련 연구의 필요성을 강조함.

•

한계점:

◦

특정 Android 애플리케이션 및 프롬프트 주입 데이터셋에 대한 실험으로 일반화의 한계 존재.

◦

공격 성공률은 LVLM 백엔드 및 특정 환경에 따라 달라질 수 있음.

◦

HG-IDA* 알고리즘의 효율성 및 최적성에 대한 추가적인 연구 필요.

◦

방어 메커니즘 개발 및 평가에 대한 연구 부족.

Made with Slashpage