Adversarial attacks on deep learning and defenses
M2, CentraleSupélec, 2024
I am the coordinator of this module since 2024.
This module is an introductory course on the use of AI for cybersecurity and the attacks that can be carried out against AI-based methods. The course does not seek to be exhaustive on all AI methods: on the contrary, it focuses on attacks against these systems, particularly those that target neural networks and degrade classification performance, which could impact intrusion detection or malware analysis systems. In addition, the course aims to describe state-of-the-art methods for implementing defense mechanisms in neural networks. This part of the course will be approached through these research aspects by researchers in the field. A practical part will enable students to implement attacks and evaluate the effectiveness of the defenses.
In this course, I wrote the lecture about adversarial attacks against deep neural networks.