Explainable AI for Cybersecurity

✍ Scribed by Zhixin Pan, Prabhat Mishra

• Publisher: Springer
• Year: 2023
• Tongue: English
• Leaves: 249
• Edition: 1st ed. 2023
• Category: Library

✦ Synopsis

This book provides a comprehensive overview of security vulnerabilities and state-of-the-art countermeasures using explainable artificial intelligence (AI). Specifically, it describes how explainable AI can be effectively used for detection and mitigation of hardware vulnerabilities (e.g., hardware Trojans) as well as software attacks (e.g., malware and ransomware). It provides insights into the security threats towards machine learning models and presents effective countermeasures. It also explores hardware acceleration of explainable AI algorithms. The reader will be able to comprehend a complete picture of cybersecurity challenges and how to detect them using explainable AI. This book serves as a single source of reference for students, researchers, engineers, and practitioners for designing secure and trustworthy systems.

✦ Table of Contents

Preface
Acknowledgements
Contents
Acronyms
Part I Introduction
Cybersecurity Landscape for Computer Systems
1 Introduction
2 Cybersecurity Vulnerabilities
2.1 Hardware Vulnerabilities
2.1.1 Malicious Implants (Hardware Trojans)
2.1.2 Supply Chain Vulnerability
2.1.3 Reverse Engineering
2.1.4 Side-Channel Leakage
2.2 Software Vulnerabilities
2.2.1 Malware Attacks
2.2.2 Ransomware Attacks
2.2.3 Spectre and Meltdown Attacks
2.3 Malicious Attacks on Machine Learning Models
2.3.1 Adversarial Attacks
2.3.2 AI Trojan Attacks
3 Detection of Security Vulnerabilities
3.1 Detection of Malicious Hardware Attacks
3.1.1 Simulation-Based Validation Using Machine Learning
3.1.2 Side-Channel Analysis Using Machine Learning
3.1.3 Heuristic Analysis Using Machine Learning
3.2 Detection of Malicious Software Attacks
3.2.1 Detection of Malware Attacks
3.2.2 Detection of Ransomware Attacks
3.2.3 Detection of Spectre and Meltdown Attacks
4 Summary
References
Explainable Artificial Intelligence
1 Introduction
2 Machine Learning Models
2.1 Support Vector Machine
2.2 Multi-Layer Perceptron
2.3 Decision Tree
2.4 Random Forest
2.5 Linear Regression
2.6 Deep Neural Network
2.7 Convolution Neural Network
2.8 Recurrent Neural Network
2.9 Long Short-Term Memory
2.10 Reinforcement Learning
2.11 Boosting
2.12 Naive Bayes
2.13 Zero-Shot Learning
3 Explainable Artificial Intelligence
3.1 Local Interpretability
3.2 Knowledge Extraction
3.3 Saliency Maps
3.4 Integrated Gradients
3.5 Shapley Value Analysis
3.6 Layer-Wise Relevance Propagation
4 Summary
References
Part II Detection of Software Vulnerabilities
Malware Detection Using Explainable AI
1 Introduction
2 Background and Related Work
2.1 Malware Detection Challenges
2.2 Why Explainable AI for Malware Detection?
3 Malware Detection Using Explainable Machine Learning
3.1 Model Training
3.2 Perturbation and Outlier Elimination
3.3 Linear Regression
3.4 Outcome Interpretation
4 Experiments
4.1 Experimental Platform
4.2 Malware and Benign Benchmarks
4.3 Data Acquisition
4.4 RNN Classifier
4.5 Evaluation: Accuracy
4.6 Evaluation: Outcome Interpretation
5 Summary
References
Spectre and Meltdown Detection Using Explainable AI
1 Introduction
1.1 Threat Model
1.2 Motivation
2 Background and Related Work
2.1 Spectre and Meltdown Attacks
2.2 Explainable Machine Learning
3 Detection of Spectre and Meltdown Attacks
3.1 Data Collection
3.2 Model Training
3.2.1 LSTM-Based Model Training
3.2.2 Ensemble Boosting
3.3 Result Interpretation
3.3.1 Explainability Using Model Distillation
3.3.2 Explainability Using Shapely Values
3.4 Data Augmentation
4 Experiments
4.1 Experimental Setup
4.2 Comparison with Existing Spectre Detection Methods
4.3 Comparison with Existing Meltdown Detection Methods
4.4 Comparison with Existing Mitigation Techniques
4.5 Stability Analysis
4.6 Explainability Analysis
4.7 Efficiency Analysis
5 Summary
References
Part III Detection of Hardware Vulnerabilities
Hardware Trojan Detection Using Reinforcement Learning
1 Introduction
2 Background and Related Work
2.1 Logic Testing for Hardware Trojan Detection
2.2 Reinforcement Learning
3 Test Generation Using Reinforcement Learning
3.1 Identification of Rare Nodes
3.2 Testability Analysis
3.3 Utilization of Reinforcement Learning
4 Experiments
4.1 Experimental Setup
4.2 Results on Trigger Coverage
4.3 Results on Test Generation Time
5 Summary
References
Hardware Trojan Detection Using Side-Channel Analysis
1 Introduction
2 Background and Motivation
2.1 Background: Reinforcement Learning
2.2 Motivation: Delay-Based Side-Channel Analysis
3 Reinforcement Learning-Based Path Delay Analysis
3.1 Overview
3.2 Generation of Initial Vectors
3.3 Generation of Succeeding Vectors
4 Experiments
4.1 Experimental Setup
4.2 Evaluation Results
5 Summary
References
Hardware Trojan Detection Using Shapley Ensemble Boosting
1 Introduction
2 Background and Related Work
2.1 Related Work for Hardware Trojan Detection
2.2 Ensemble Boosting
2.3 Shapley Values
3 Shapley Ensemble Boosting for Hardware Trojan Detection
3.1 Data Sampling
3.2 Model Training
3.3 Shapley Analysis
3.4 Weight Adjustment
3.5 Ensemble Prediction
4 Experiments
4.1 Experimental Setup
4.2 HT Detection Performance
4.3 Explainability Analysis
4.4 Efficiency Analysis
4.5 Robustness Analysis
5 Summary
References
Part IV Mitigation of AI Vulnerabilities
Mitigation of Adversarial Machine Learning
1 Introduction
2 Background and Preliminaries
2.1 Attacks on Neural Networks
2.2 Spectral Normalization
3 Spectral Normalization to Defend Against Adversarial Attacks
3.1 Layer Separation
3.2 Fourier Transform
3.3 Activation Functions
3.4 Complexity Analysis
4 Experiments
4.1 Experimental Setup
4.2 Case Study: MNIST Benchmark
4.3 Case Study: ImageNet Benchmark
5 Summary
References
AI Trojan Attacks and Countermeasures
1 Introduction
2 Background and Related Work
3 Backdoor Attack with AI Trojans
3.1 Feature Extraction
3.2 Normal Training
3.3 Backdoor Training
3.4 Trojan Injection
4 Defenses Against AI Trojans
4.1 Pruning
4.2 Bayesian Neural Networks
4.3 Neural Cleanse
4.4 Artificial Brain Stimulation (ABS)
4.5 STRIP
5 Experiments
5.1 Experimental Setup
5.2 Comparison of Attack Performance
5.3 Overhead Analysis
5.4 Robustness Against STRIP-Based Defense
6 Summary
References
Part V Acceleration of Explainable AI
Hardware Acceleration of Explainable AI
1 Introduction
1.1 Graphics Processing Unit
1.2 Field Programmable Gate Array
2 FPGA-Based Acceleration of Heatmap Visualization
3 FPGA-Based Acceleration of Saliency Map
4 GPU-Based Acceleration of Shapley Value Analysis
4.1 Summary
References
Explainable AI Acceleration Using Tensor Processing Units
1 Introduction
2 Tensor Processing Units
3 Hardware Acceleration of Explainable AI
3.1 Task Transformation
3.2 Data Decomposition in Fourier Transform
4 Experiments
4.1 Experimental Setup
4.2 Comparison of Accuracy and Classification Time
4.3 Comparison of Energy Efficiency
5 Summary
References
Part VI Conclusion
The Future of AI-Enabled Cybersecurity
1 Introduction
2 Summary
2.1 Introduction to Cybersecurity and Explainable AI
2.1.1 Detection of Software Vulnerabilities
2.2 Detection of Hardware Vulnerabilities
2.3 Mitigation of AI Vulnerabilities
2.4 Acceleration of Explainable AI
3 Future Directions
3.1 Automatic Implementation of Secure Systems
3.2 Detection of Malicious Implants
3.3 Detection of Ransomware Attacks
3.4 Automatic Data Augmentation
References
Index