Attacking Machine Learning Methods Used for Detection of Cyber Attack

Explore the vulnerabilities of machine learning methods used for cyber attack detection in this 45-minute conference talk. Delve into the increasing complexity of communication networks and the rising interest of attackers in gaining access to information. Examine why machine learning methods are proposed for effective cyber attack detection, including their scalability, speed, and ability to protect against unknown threats. Discover various inherent properties of machine learning methods that allow attackers to bypass detection systems. Learn about specific attacks on these methods and their impact on detection performance. Gain insights into current open problems in machine learning security and cyber attack detection. Understand topics such as model stealing, model inversion, evasion attacks, and defensive strategies like distillation and effective defenses. Conclude with a discussion on evolution scenarios, cross-transfer ability, and limitations in the field.