Computer Security - ESORICS 2024

Name: Computer Security - ESORICS 2024 | 29th European Symposium on Research in Computer Security, Bydgoszcz, Poland, September 16-20, 2024, Proceedings, Part I
Brand: Springer
Price: 69.54 EUR
Availability: OnlineOnly

29th European Symposium on Research in Computer Security, Bydgoszcz, Poland, September 16-20, 2024, Proceedings, Part I

Joaquin Garcia-Alfaro Rafal Kozik Michal Choras Sokratis Katsikas(Herausgeber*in)

Springer (Verlag)

Erschienen am 4. September 2024

XVI, 396 Seiten

E-Book

PDF mit Wasserzeichen-DRM

Systemvoraussetzungen

978-3-031-70879-4 (ISBN)

69,54 €inkl. 7% MwSt.

Systemvoraussetzungen

für PDF mit Wasserzeichen-DRM

E-Book Einzellizenz

Als Download verfügbar

Beschreibung

Weitere Details

Weitere Ausgaben

Inhalt

Intro
Preface
Organization
Contents - Part I
Security and Machine Learning
Attesting Distributional Properties of Training Data for Machine Learning
1 Introduction
2 Background
3 Problem Statement
4 Distributional Property Attestation Mechanisms
5 Experimental Setup
6 Experimental Evaluation
6.1 Inference-Based Attestation
6.2 Cryptographic Attestation
6.3 Hybrid Attestation
7 Related Work
8 Discussions
A Details for Cryptographic Attestation
References
Towards Detection-Recovery Strategy for Robust Decentralized Matrix Factorization
1 Introduction
2 Background and Related Work
2.1 Decentralized Matrix Factorization
2.2 Threats and Remedies in Distributed Learning
3 The Vulnerability of DMF
3.1 Threat Model
3.2 The Tampering Attack on DMF
4 Our Approach
4.1 The Decentralized Detection
4.2 The Recovery Strategy
4.3 Comprehensive Framework
5 Experiment
5.1 Experimental Setup
5.2 The Threat of the Tampering Attack
5.3 Effective Defense with the Detection-Recovery Strategy
5.4 Adaptive Attack
5.5 More Results
6 Conclusion and Discussion
A Technical Proofs
References
Bayesian Learned Models Can Detect Adversarial Malware for Free
1 Introduction
2 Background and Related Work
3 Problem Definition
3.1 Threat Model
3.2 Adversarial Malware Attacks
4 Measuring Uncertainty
4.1 Bayesian Machine Learning for Malware Detection
4.2 Uncertainty Measures
5 Experiments and Results
5.1 Experimental Setup
5.2 Clean Performance (No Attacks) in Android Domain
5.3 Robustness Against Problem-Space Adversarial Android Malware
5.4 Robustness Against Feature-Space Adversarial Android Malware
5.5 Generalization to PDF Malware
5.6 Generalization to Windows PE Files
6 Identifying Concept Drift
7 Model Parameter Diversity Measures
8 Threat to Validity
9 Conclusion
References
Resilience of Voice Assistants to Synthetic Speech
1 Introduction
2 Voice Assistants
3 Related Work
3.1 Deepfake Speech Synthesis
3.2 Spofing Attacks on Biometrics Systems
3.3 Spoofing Voice Assistants
4 Experiments
4.1 Used Speech Synthesisers
4.2 Environment Description
4.3 Details of the Setup
5 Experimental Evaluation
6 Threat Analysis
7 Discussion
7.1 Observations
7.2 Mitigation Methods
8 Conclusions
References
Have You Poisoned My Data? Defending Neural Networks Against Data Poisoning
1 Introduction
2 Background
2.1 Feature Collision
2.2 Convex Polytope and Bullseye Polytope
2.3 Gradient Matching
3 System and Threat Models
3.1 System Model
3.2 Threat Model
4 Our Approach
4.1 Formal Description of the Approach
5 Experimental Setup
5.1 Dataset
5.2 Poison Generation Algorithms and Defenses
6 Evaluation
6.1 Poisons vs Clean Samples: A Characteristic Vector Perspective
6.2 Poison Detection
7 Related Works
8 Conclusions and Future Work
A Implementation Details
B Additional Experimental Results
References
Jatmo: Prompt Injection Defense by Task-Specific Finetuning
1 Introduction
2 Background
2.1 LLM-Integrated Applications
2.2 Prompt Injections
2.3 Examples
3 Related Works
3.1 Types of Attacks
3.2 Pitfalls of Traditional Defenses
4 Jatmo
4.1 Synthetic Input Generation
5 Results
5.1 Experimental Methodology
5.2 Main Results
5.3 Training with Less Data
5.4 Synthetic Dataset Generation
6 Discussion
7 Summary
A Appendix
A.1 Detailed Task Parameters
References
PointAPA: Towards Availability Poisoning Attacks in 3D Point Clouds
1 Introduction
2 Related Work
2.1 Adversarial Attacks of 3D Point Clouds
2.2 Backdoor Attacks of 3D Point Clouds
2.3 Availability Poisoning Attacks in 2D Images
3 Methodology
3.1 Threat Model
3.2 Motivation and Challenges
3.3 Inspiration and Exploration
3.4 PointAPA: Point Cloud Availability Poisoning Attack
3.5 Why Does PointAPA Work?
4 Experiments
4.1 Experimental Settings
4.2 Evaluation on PointAPA
4.3 Evaluation Under Overlapped Rotation Angles
4.4 Robustness to Defense Schemes
4.5 Hyper-parameter Analysis
5 Conclusion
A Appendix
References
ECLIPSE: Expunging Clean-Label Indiscriminate Poisons via Sparse Diffusion Purification
1 Introduction
2 Related Work
2.1 Clean-Label Indiscriminate Poisoning Attacks
2.2 Defenses Against Poisoning Attacks
3 Methodology
3.1 Threat Model
3.2 Motivation for Studying Defenses Against CLBPAs
3.3 Key Intuition and Theoretical Insight
3.4 Challenges and Approaches
3.5 Our Design for ECLIPSE
4 Experiments
4.1 Experimental Settings
4.2 Evaluation of ECLIPSE
4.3 Purification Visual Effect
4.4 Resistance to Potential Adaptive Attacks
4.5 Hyper-Parameter Analysis
4.6 Ablation Study
4.7 Analysis of ECLIPSE
5 Conclusion and Limitation
A Appendix
References
MAG-JAM: Jamming Detection via Magnetic Emissions
1 Introduction
2 MAG-JAM Overview, Scenario and Adversary Model
2.1 MAG-JAM Overview
2.2 Scenario and Adversary Model
3 Jamming Detection Using Magnetic Sensor
3.1 DRV425 Magnetic Sensor Setup
3.2 Magnetic Sensor Results
3.3 Early Jamming Detection
4 MAG-JAM Evaluation
4.1 Experimental Setup - Magnetic Probe
4.2 Magnetic Emissions Collection Using the Magnetic Probe
4.3 Dataset Description
4.4 Features Extraction
4.5 Jamming Detection Using Autoencoder
5 Discussion
6 Related Work
7 Conclusion
References
Fake or Compromised? Making Sense of Malicious Clients in Federated Learning
1 Introduction
2 Types of Byzantine-Robust Aggregation Rules
3 Distinguishing Fake And Compromised Adversary Models
3.1 Adversary with Fake Clients
3.2 Adversary with Compromised Clients
4 Our Proposed Hybrid Adversary Model
4.1 Comparing the Costs of Different Adversaries
5 Experimental Setup
5.1 Datasets and Hyperparameters
5.2 Evaluation Metric
5.3 Generating Synthetic Data Using DDPM
6 Experiments
6.1 Attacking Agnostic Robust AGRs
6.2 Attacking Adaptive Robust AGRs
7 Conclusions
A Auxiliary Results of Model Poisoning Attacks Against Aware AGRs
References
Beyond Words: Stylometric Analysis for Detecting AI Manipulation on Social Media
1 Introduction
2 Related Work
2.1 Pervasiveness and Influence of Social Bots
2.2 Evaluation and Detection of Social Bots and AI-Text
3 Study Design
3.1 Data Generation and Preparation
3.2 Stylometric Analysis
3.3 Analysis Methods
4 Results
5 Threats to Validity
6 Conclusions
References
FSSiBNN: FSS-Based Secure Binarized Neural Network Inference with Free Bitwidth Conversion
1 Introduction
1.1 Related Work
1.2 Our Contributions
2 Preliminaries
2.1 Binarized Neural Networks
2.2 Additive Secret Sharing
2.3 Function Secret Sharing
3 Secure BNN Inference Framework
3.1 The FSSiBNN Overview
3.2 Bitwidth-Reduced Parameter Encoding Scheme with Free Bitwidth Conversion
3.3 Online-Efficient Secure Non-linear BNN Layers via FSS
4 Secure BNN Inference Protocol
4.1 Secure Fully Connected and Convolutional Layers
4.2 Secure Batch Normalization and Binary Activation Layers
4.3 Secure Max Pooling Layers
5 Theoretical Analysis and Experiment
5.1 Theoretical Analysis
5.2 Experimental Results and Analysis
6 Conclusion
A Proof of Sign Function Gate in Sect.4.2
B Analysis of Computation Complexity
C Evaluation and Analysis of Inference Accuracy
References
Optimal Machine-Learning Attacks on Hybrid PUFs
1 Introduction
1.1 Problem Statement and Related Work
1.2 Contributions
1.3 Paper Organisation
2 Mathematical Representations of Hybrid PUFs
2.1 XOR Arbiter PUF
2.2 OR-AND-XOR-PUF
2.3 Homogeneous and Heterogeneous Feed-Forward XOR Arbiter PUF
2.4 Other Hybrid PUFs
2.5 State-of-Art Modelling Structures
3 Methodology
3.1 Local Minima Problem
3.2 Modelling PUFs Using Miture-of-Experts
3.3 Routine Algorithm
3.4 Proposed Transition Theorem
4 Experiments and Evaluation
4.1 Modelling Hybrid PUFs Using the Generic Model
4.2 Modelling Hybrid PUFs Using the Proposed Transition Theorem
5 Conclusion
A Transition Theorem and Proofs
A.1 OAX-PUF
B Feed-Forward PUF
References
Outside the Comfort Zone: Analysing LLM Capabilities in Software Vulnerability Detection
1 Introduction
2 Related Work
2.1 SAST-Based Vulnerability Detection
2.2 Task-Specific DL Models for Vulnerability Detection
2.3 LLM-Based Vulnerability Detection
3 Methodology
4 Experiments
4.1 Prompt Engineering and Hardware Setup
4.2 Datasets
5 Results and Discussion
6 Conclusions
References
ZeroLeak: Automated Side-Channel Patching in Source Code Using LLMs
1 Introduction
2 Background
3 Related Work
4 Threat Model and Scope
5 Methodology
5.1 Ensuring Constant-Time Execution
5.2 Mitigating Spectre-v1
6 Evaluation
6.1 Patching Spectre-v1 Gadgets
6.2 Patching a Real World Spectre-v1 Gadget
6.3 Patching Real-World Javascript Libraries for Constant-Timeness
6.4 Comparison of LLMs
7 Discussion and Limitations
8 Conclusion
A Spectre-v1 Patch in OpenSSL Generated by ZeroLeak
References
The Adversarial AI-Art: Understanding, Generation, Detection, and Benchmarking
1 Introduction
2 Background: AI-Generated Multimedia Content
3 The AdversaRIal AI-Art (ARIA) Dataset
3.1 Terminology: Real Images and AI-Generated Images
3.2 The Real Images and Annotation
3.3 AI-Art Generation
3.4 The ARIA Dataset
4 The User Study
4.1 The Design
4.2 The Results and Analysis
5 Benchmarking AI-Image Detectors
6 Discussions
7 Related Work
8 Copyright and Ethical Considerations
9 Conclusion
References
Optimizing Cyber Defense in Dynamic Active Directories Through Reinforcement Learning
1 Introduction
2 Related Work
3 Problem Description
4 Proposed Attacker-Defender Approach
4.1 Proposed AD Graph Optimization Technique
4.2 Attacker Approach: Reinforcement Learning
4.3 RL Training Facilitator: Pruning Approaches
4.4 Defender's Approach: Reinforcement Learning Assisted Evolutionary Diversity Optimization
4.5 Overall Attacker-Defender Approach
5 Experimental Results
5.1 Synthetic AD Graph Dataset
5.2 Training Parameters
5.3 Attacker-Defender Policy Training
5.4 Evaluating Attacker's Policy
5.5 Evaluating Defender's Policy
5.6 Discussion
6 Conclusion
References
CryptoLLM: Harnessing the Power of LLMs to Detect Cryptographic API Misuse
1 Introduction
2 Cryptographic API Misuse Vulnerabilities
3 Overview of CryptoLLM
4 Implementation of CryptoLLM
4.1 Code Snippet Generation
4.2 LLM Construction
5 Experiments
5.1 Data Collection
5.2 Experimental Setup
5.3 Model Optimization
6 Evaluation
6.1 Performance Comparison on the CryptoAPI-Bench Dataset
6.2 Performance Comparison on Unseen Real-World APK Files
6.3 Evaluation of Robustness on the Mutated Dataset
7 Discussion
7.1 Comparison with Existing Approaches
7.2 Generalization of Experimental Results
7.3 Lack of Ground-Truth Labeling
8 Related Work
9 Conclusion
References
GAN-GRID: A Novel Generative Attack on Smart Grid Stability Prediction
1 Introduction
2 Related Work
2.1 AI and ML for SG Stability
2.2 Adversarial Attacks
3 System and Threat Model
4 GAN-GRID: Our Proposed Adversarial Attack
4.1 GAN-GRID Model
4.2 Reference Whitebox Attacks
5 Grid Stability Prediction
5.1 ML Model Design
5.2 DL Model Design
6 Evaluation
6.1 Dataset
6.2 Baseline Evaluation
6.3 Attack Evaluation
7 Conclusions
References
Author Index

Systemvoraussetzungen

Als PDF speichern Als Link merken