Malware Detection in the AI Era: Attacks and Defenses on Machine Learning Classifiers

Endpoint Detection and Response (EDR) systems and Antivirus (AV) solutions have incorporated machine learning (ML) as core components of their decision-making processes. However, the integration of ML has introduced new vulnerabilities, rendering these systems susceptible to specific types of attacks that can weaken their effectiveness.

In this course, participants will first gain a comprehensive understanding of how machine learning models can perform the task of malware detection in both static and dynamic settings, and they will use techniques that explain their behavior.

Furthermore, we will introduce the concepts of Adversarial Machine Learning, the field of science that formalizes the presence of an adversary whose intent is the exploitation of AI models. Attendees will first learn and then execute known adversarial strategies designed to compromise ML malware classifiers under different threat models. Lastly, we will show how these attacks can be limited, by discussing recent advancements in research of defensive mechanisms.