Adversarial Machine Learning
by Jason Edwards;
English | 2026 | ISBN: 1394402031 | 396 pages | True PDF EPUB | 6.13 MB
Enables readers to understand the full lifecycle of adversarial machine learning (AML) and how AI models can be compromised
Adversarial Machine Learning is a definitive guide to one of the most urgent challenges in artificial intelligence today: how to secure machine learning systems against adversarial threats.
This book explores the full lifecycle of adversarial machine learning (AML), providing a structured, real-world understanding of how AI models can be compromised―and what can be done about it.
The book walks readers through the different phases of the machine learning pipeline, showing how attacks emerge during training, deployment, and inference. It breaks down adversarial threats into clear categories based on attacker goals―whether to disrupt system availability, tamper with outputs, or leak private information. With clarity and technical rigor, it dissects the tools, knowledge, and access attackers need to exploit AI systems.
In addition to diagnosing threats, the book provides a robust overview of defense strategies―from adversarial training and certified defenses to privacy-preserving machine learning and risk-aware system design. Each defense is discussed alongside its limitations, trade-offs, and real-world applicability.
Readers will gain a comprehensive view of today???s most dangerous attack methods including:
Evasion attacks that manipulate inputs to deceive AI predictions
Poisoning attacks that corrupt training data or model updates
Backdoor and trojan attacks that embed malicious triggers
Privacy attacks that reveal sensitive data through model interaction and prompt injection
Generative AI attacks that exploit the new wave of large language models
Blending technical depth with practical insight, Adversarial Machine Learning equips developers, security engineers, and AI decision-makers with the knowledge they need to understand the adversarial landscape and defend their systems with confidence.
Links are Interchangeable - Single Extraction
