|
Abstract:
|
As more and more security data are collected, machine learning techniques become an essential tool for real-world security applications. One of the most important differences between cyber security and many other applications is the existence of malicious adversaries that actively adapt their behavior to make the existing learning models ineffective. Unfortunately, traditional learning techniques are insufficient to handle such adversarial problems directly. The adversaries adapt to the defender's reactions, and learning algorithms constructed based on the current training dataset degrades quickly. Based on a game theoretic framework to model the sequential actions of the adversaries and the defender, we develop adversarial classification and adversarial clustering methods to defend against active adversaries. An adversarial attack against deep neural networks is introduced in this talk too.
|
