/
/
Daily Arxiv
Daily Arxiv
世界中で発行される人工知能関連の論文をまとめるページです。
このページはGoogle Geminiを活用して要約し、非営利で運営しています。
論文の著作権は著者および関連機関にあり、共有する際は出典を明記してください。
Arbitrary Precision Printed Ternary Neural Networks with Holistic Evolutionary Approximation
Invited Paper: Feature-to-Classifier Co-Design for Mixed-Signal Smart Flexible Wearables for Healthcare at the Extreme Edge
Robustness is Important: Limitations of LLMs for Data Fitting
CMPhysBench: A Benchmark for Evaluating Large Language Models in Condensed Matter Physics
BudgetThinker: Empowering Budget-aware LLM Reasoning with Control Tokens
CE-RS-SBCIT A Novel Channel Enhanced Hybrid CNN Transformer with Residual, Spatial, and Boundary-Aware Learning for Brain Tumor MRI Analysis
PlantVillageVQA: A Visual Question Answering Dataset for Benchmarking Vision-Language Models in Plant Science
THEME: Enhancing Thematic Investing with Semantic Stock Representations and Temporal Dynamics
Trust but Verify! A Survey on Verification Design for Test-time Scaling
Quantized Neural Networks for Microcontrollers: A Comprehensive Review of Methods, Platforms, and Applications
Documenting Deployment with Fabric: A Repository of Real-World AI Governance
Atom-Searcher: Enhancing Agentic Deep Research via Fine-Grained Atomic Thought Reward
Region-Level Context-Aware Multimodal Understanding
ETTRL: Balancing Exploration and Exploitation in LLM Test-Time Reinforcement Learning Via Entropy Mechanism
Mask & Match: Learning to Recognize Handwritten Math with Self-Supervised Attention
Adaptive Duration Model for Text Speech Alignment
SKA-Bench: A Fine-Grained Benchmark for Evaluating Structured Knowledge Understanding of LLMs
Time-RA: Towards Time Series Reasoning for Anomaly with LLM Feedback
Dually Hierarchical Drift Adaptation for Online Configuration Performance Learning
Single Domain Generalization for Multimodal Cross-Cancer Prognosis via Dirac Rebalancer and Distribution Entanglement
Interpretable Mnemonic Generation for Kanji Learning via Expectation-Maximization
Quantifying Fairness in LLMs Beyond Tokens: A Semantic and Statistical Perspective
BASE-Q: Bias and Asymmetric Scaling Enhanced Rotational Quantization for Large Language Models
Scientifically-Interpretable Reasoning Network (ScIReN): Discovering Hidden Relationships in the Carbon Cycle and Beyond
A Hybrid Artificial Intelligence Method for Estimating Flicker in Power Systems
Beyond Frequency: The Role of Redundancy in Large Language Model Memorization
TrueGL: A Truthful, Reliable, and Unified Engine for Grounded Learning in Full-Stack Search
Unified Path Planner with Adaptive Safety and Optimality
FedSEA-LLaMA: A Secure, Efficient and Adaptive Federated Splitting Framework for Large Language Models
WebInject: Prompt Injection Attack to Web Agents
Towards Embodiment Scaling Laws in Robot Locomotion
SPIN-ODE: Stiff Physics-Informed Neural ODE for Chemical Reaction Rate Estimation
DDaTR: Dynamic Difference-aware Temporal Residual Network for Longitudinal Radiology Report Generation
Latent Adaptive Planner for Dynamic Manipulation
MAC-Tuning: LLM Multi-Compositional Problem Reasoning with Enhanced Knowledge Boundary Awareness
SAGA: A Security Architecture for Governing AI Agentic Systems
Towards Understanding Camera Motions in Any Video
Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction
DeepTrans: Deep Reasoning Translation via Reinforcement Learning
A Hybrid Fully Convolutional CNN-Transformer Model for Inherently Interpretable Disease Detection from Retinal Fundus Images
Decentralized Domain Generalization with Style Sharing: Formal Model and Convergence Analysis
FROG: Fair Removal on Graphs
DPImageBench: A Unified Benchmark for Differentially Private Image Synthesis
LLM Test Generation via Iterative Hybrid Program Analysis
Toxicity Begets Toxicity: Unraveling Conversational Chains in Political Podcasts
Retrieval-Augmented Machine Translation with Unstructured Knowledge
ROSE: A Reward-Oriented Data Selection Framework for LLM Task-Specific Instruction Tuning
RevPRAG: Revealing Poisoning Attacks in Retrieval-Augmented Generation through LLM Activation Analysis
Categorical Data Clustering via Value Order Estimated Distance Metric Learning
Guiding a diffusion model using sliding windows
A Collaborative Content Moderation Framework for Toxicity Detection based on Conformalized Estimates of Annotation Disagreement
Mamba State-Space Models Are Lyapunov-Stable Learners
Alice's Adventures in a Differentiable Wonderland - Volume I, A Tour of the Land
COBRA-PPM: A Causal Bayesian Reasoning Architecture Using Probabilistic Programming for Robot Manipulation Under Uncertainty
Large Intestine 3D Shape Refinement Using Point Diffusion Models for Digital Phantom Generation
What Breaks Knowledge Graph based RAG? Empirical Insights into Reasoning under Incomplete Knowledge
QHackBench: Benchmarking Large Language Models for Quantum Code Generation Using PennyLane Hackathon Challenges
AI Simulation by Digital Twins: Systematic Survey, Reference Framework, and Mapping to a Standardized Architecture
Compression versus Accuracy: A Hierarchy of Lifted Models
TrustGeoGen: Formal-Verified Data Engine for Trustworthy Multi-modal Geometric Problem Solving
Evaluating Knowledge Graph Based Retrieval Augmented Generation Methods under Knowledge Incompleteness
Transforming Wearable Data into Personal Health Insights using Large Language Model Agents
Policy Expansion for Bridging Offline-to-Online Reinforcement Learning
The Demon is in Ambiguity: Revisiting Situation Recognition with Single Positive Multi-Label Learning
DynaMark: A Reinforcement Learning Framework for Dynamic Watermarking in Industrial Machine Tool Controllers
TMUAD: Enhancing Logical Capabilities in Unified Anomaly Detection Models with a Text Memory Bank
MoE-Health: A Mixture of Experts Framework for Robust Multimodal Healthcare Prediction
Going over Fine Web with a Fine-Tooth Comb: Technical Report of Indexing Fine Web for Problematic Content Search and Retrieval
PiCSAR: Probabilistic Confidence Selection And Ranking
Benchmarking GPT-5 in Radiation Oncology: Measurable Gains, but Persistent Need for Expert Oversight
Unsupervised Video Continual Learning via Non-Parametric Deep Embedded Clustering
Reasoning-Intensive Regression
Neural Network Acceleration on MPSoC board: Integrating SLAC's SNL, Rogue Software and Auto-SNL
Developer Insights into Designing AI-Based Computer Perception Tools
CAD2DMD-SET: Synthetic Generation Tool of Digital Measurement Device CAD Model Datasets for fine-tuning Large Vision-Language Models
OptMark: Robust Multi-bit Diffusion Watermarking via Inference Time Optimization
Entropy-Based Non-Invasive Reliability Monitoring of Convolutional Neural Networks
Why Stop at Words? Unveiling the Bigger Picture スルー Line-Level OCR
Harnessing IoT and Generative AI for Weather-Adaptive Learning in Climate Resilience Education
QZhou-Embedding Technical Report
Physics-Informed Spectral Modeling for Hyperspectral Imaging
Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning
A Survey on Current Trends and Recent Advances in Text Anonymization
NSPDI-SNN: An efficient lightweight SNN based on nonlinear synaptic pruning and dendritic integration
Limitations of Physics-Informed Neural Networks: a Study on Smart Grid Surrogation
EZ-Sort: Efficient Pairwise Comparison via Zero-Shot CLIP-Based Pre-Ordering and Human-in-the-Loop Sorting
What Data is Really Necessary? A Feasibility Study of Inference Data Minimization for Recommender Systems
Complete Gaussian Splats from a Single Image with Denoising Diffusion Models
On the Hardness of Learning GNN-based SAT Solvers: The Role of Graph Ricci Curvature
ELV-Halluc: Benchmarking Semantic Aggregation Hallucinations in Long Video Understanding
Priors Matter: Addressing Misspecification in Bayesian Deep Q-Learning
HSFN: Hierarchical Selection for Fake News Detection building Heterogeneous Ensemble
Igniting Creative Writing in Small Language Models: LLM-as-a-Judge versus Multi-Agent Refined Rewards
Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration
Diffusion-based Multi-modal Synergy Interest Network for Click-through Rate Prediction
MedShift: Implicit Conditional Transport for X-Ray Domain Adaptation
The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Management
Med-RewardBench: Benchmarking Reward Models and Judges for Medical Multimodal Large Language Models
Benchmarking the State of Networks with a Low-Cost Method Based on Reservoir Computing
DRASP: A Dual-Resolution Attentive Statistics Pooling Framework for Automatic MOS Prediction
Load more
Prefill-level Jailbreak: A Black-Box Risk Analysis of Large Language Models
Created by
Haebom
作者
Yakai Li, Jiekang Hu, Weiduan Sang, Luping Ma, Dongsheng Nie, Weijuan Zhang, Aimin Yu, Yi Su, Qingjia Huang, Qihang Zhou
概要
本論文は、大規模言語モデル(LLM)のセキュリティ脅威の1つである在日ブレイク攻撃について、既存の研究で主に取り上げられていたプロンプトレベルの攻撃ではなく、ユーザー制御応答の事前充填機能を悪用する攻撃に焦点を当てて研究した結果を提示します。プリフィルにより、攻撃者はモデル出力の先頭を操作して、説得ベースの攻撃からモデル状態直接操作に攻撃パラダイムを切り替えます。 14のLLMを対象にブラックボックスセキュリティ分析を行い、事前充填レベルの在日ブレーキ攻撃を分類し、その効果を評価しました。実験の結果、適応方法を使用した攻撃は、複数のモデルで99%を超える成功率を達成し、トークンレベルの確率分析により、初期状態操作によって拒否から協力への最初のトークン確率の変化が発生することが確認されました。また、事前充填レベルの在日ブレーキ攻撃は、既存のプロンプトレベル攻撃の成功率を10~15%p向上させる効果的な増強剤として機能することを示しています。いくつかの防御戦略評価の結果、既存のコンテンツフィルタは制限的な保護効果のみを提供し、プロンプトとプリフィルとの間の操作的関係に焦点を当てた検出方法がより効果的であることがわかりました。結論として、現在のLLM安全アライメントの脆弱性を明らかにし、将来の安全訓練で事前充填攻撃領域を解決する必要性を強調する。
Takeaways、Limitations
•
Takeaways:
◦
ユーザー制御応答プレフィル機能を使用した新しいタイプの在日ブレーキ攻撃の存在と重大性を明らかにします。
◦
プリフィル攻撃が既存のプロンプトベースの攻撃を増幅できることを示しています。
◦
既存のコンテンツフィルタの限界を明らかにし、プロンプトとプレフィルの間の関係に基づく新しい検出方法の必要性を提示する。
◦
LLMの安全性を向上させるための研究方向の提示(prefill攻撃対応)。
•
Limitations:
◦
分析対象モデルの種類と数の制限(14モデル)。
◦
提案された検出方法の一般化の可能性と実際の環境適用に関するさらなる研究の必要性
◦
さまざまな種類のプレフィル攻撃の包括的な分析が不足する可能性があります。
PDFを見る
Made with Slashpage