A survey on adversarial machine learning: Attacks, defenses, real-world applications, and future research directions
The rapid proliferation of machine learning (ML) systems across critical domains has heightened concerns about their susceptibility to adversarial threats. This survey offers a comprehensive overview of adversarial machine learning, synthesizing a broad body of research encompassing attack methodologies, defense strategies, and real-world applications. We present a systematic taxonomy of adversarial threats spanning the ML lifecycle, including training-time attacks such as data poisoning and backdoor insertion, as well as inference-time attacks such as evasion, model extraction, and privacy leakage. We examine a wide range of defense mechanisms, including proactive approaches (e.g., adversarial training and input sanitization), detection-based techniques that leverage statistical or behavioral signatures, and reactive strategies such as model patching and ensemble learning. We further discuss recent advances in privacy-preserving machine learning, including differential privacy, federated learning, and secure aggregation. Through real-world case studies in domains such as computer vision, natural language processing, autonomous systems, and healthcare, we highlight persistent vulnerabilities and practical challenges. Finally, we outline critical open problems and promising directions for future research. This work consolidates current understanding and serves as a foundational reference for enhancing the security and robustness of machine learning systems.
