Cyber Security Meets Machine Learning

Machine learning boosts the capabilities of security solutions in the modern cyber environment. However, there are also security concerns associated with machine learning models and approaches: the vulnerability of machine learning models to adversarial attacks is a fatal flaw in the artificial intelligence technologies, and the privacy of the data used in the training and testing periods is also causing increasing concern among users. This book reviews the latest research in the area, including effective applications of machine learning methods in cybersecurity solutions and the urgent security risks related to the machine learning models. The book is divided into three parts: Cyber Security Based on Machine Learning; Security in Machine Learning Methods and Systems; and Security and Privacy in Outsourced Machine Learning. Addressing hot topics in cybersecurity and written by leading researchers in the field, the book features self-contained chapters to allow readers to select topics that are relevant to their needs. It is a valuable resource for all those interested in cybersecurity and robust machine learning, including graduate students and academic and industrial researchers, wanting to gain insights into cutting-edge research topics, as well as related tools and inspiring innovations. Preface 5 Contents 8 IoT Attacks and Malware 9 1 Introduction 9 2 Background 12 2.1 Cybersecurity Kill Chains 12 2.2 Major IoT Security Concerns 13 3 Attack Classification 14 3.1 Passive/Information Stealing Attacks 14 3.2 Service Degradation Attacks 17 3.3 DDoS Attacks 20 4 IoT Malware Analysis and Classification 27 5 AI-Based IDS Solutions 29 6 Conclusion 30 References 31 Machine Learning-Based Online Source Identification for Image Forensics 34 1 Introduction 34 2 Related Work 36 2.1 Features Engineering for Image Source Identification 36 2.2 Statistical Learning-Based Image Source Identification 37 3 Proposed Scheme: OSIU 38 3.1 Unknown Sample Triage 39 3.2 Unknown Image Discovery 44 3.3 (K+1)-class Classification 49 4 Experiments and Results 49 4.1 Dataset and Experiment Settings 49 4.2 Features 51 4.3 Evaluation Metrics 52 4.4 Performance of Triaging Unknown Samples 52 4.5 Performance of OSIU 57 5 Conclusion 61 References 61 Reinforcement Learning Based Communication Security for Unmanned Aerial Vehicles 64 1 Introduction 64 2 Communication Security for Unmanned Aerial Vehicles 66 2.1 UAV Communication Model 66 2.2 Attack Model 67 3 Reinforcement Learning Based UAV Communication Security 67 3.1 Reinforcement Learning Based Anti-Jamming Communications 67 3.2 Reinforcement Learning Based UAV Communications Against Smart Attacks 72 4 UAV Secure Communication Game 78 4.1 Game Model 78 4.2 Nash Equilibrium of the Game 82 5 Related Work 86 5.1 General Anti-jamming Policies in UAV-Aided Communication 86 5.2 Reinforcement Learning in Anti-jamming Communication 86 5.3 Game Theory in Anti-jamming Communication 87 6 Conclusion 88 References 89 Visual Analysis of Adversarial Examples in Machine Learning 91 1 Introduction 91 2 Adversarial Examples 93 3 Generation of Adversarial Examples 94 4 Properties of Adversarial Examples 95 5 Distinguishing Adversarial Examples 96 6 Robustness of Models 98 7 Challenges and Research Directions 98 8 Conclusion 99 References 99 Adversarial Attacks Against Deep Learning-Based Speech Recognition Systems 105 1 Introduction 105 2 Background and Related Work 107 2.1 Speech Recognition 107 2.2 Adversarial Examples 107 2.3 Related Work 108 3 Overview 109 3.1 Motivation 109 3.2 Technical Challenges 110 4 White-Box Attack 110 4.1 Threat Model of White-Box Attack 111 4.2 The Detail Decoding Process of Kaldi 111 4.3 Gradient Descent to Craft Audio Clip 113 4.4 Practical Adversarial Attack Against White-Box Model 114 4.5 Experiment Setup of CommanderSong Attack 115 4.6 Evaluation of CommanderSong Attack 116 5 Black-Box Attack 118 5.1 Threat Model of Black-Box Attack 119 5.2 Transferability Based Approach 119 5.3 Local Model Approximation Approach 120 5.4 Alternate Models Based Generation Approach 121 5.5 Experiment Setup of Devil's Whisper Attack 122 5.6 Evaluation of Devil's Whisper Attack 123 6 Defense 125 7 Conclusion 126 Appendix 127 References 132 A Survey on Secure Outsourced Deep Learning 134 1 Introduction 134 2 Deep Learning 136 2.1 Brief Survey on Deep Learning 136 2.2 Architecture of Deep Learning 138 2.3 Main Computation in Deep Learning 140 3 Outsourced Computation 142 3.1 Brief Survey on Outsourced Computation 142 3.2 System Model 144 3.3 Security Requirements 144 4 Outsourced Deep Learning 145 4.1 Brief Review on Outsourced Deep Learning 146 4.2 Privacy Concerns in Outsourced Deep Learning 147 4.3 Privacy-Preserving Techniques for Outsourced Deep Learning 148 4.4 Taxonomy Standard 151 4.5 Privacy-Preserving Training Outsourcing 151 4.6 Privacy-Preserving Inference Outsourcing 159 5 Conclusion and Future Research Perspectives 162 References 164