Winning the battle for secure ML

Abstract

This book provides a comprehensive yet methodical understanding of securing today's AI systems. It covers vulnerabilities throughout the complete machine learning life cycle from data collection, to training, and deployment and inference, as well as presents practical methods for mitigating the most harmful threats. By integrating theoretical foundations, practical case studies, and recent research, the book covers essential topics including threat modelling, adversarial attacks, poisoning attacks, and privacy breaches.