Fr

Research Seminar

The IMC2 seminar series kicked off strongly with a presentation on a key trend in cybersecurity: adversarial attacks in artificial intelligence.

A big thank you to our guest speaker, Adel Abusitta, as well as to all participants joining us live from HEC Montréal and Université du Québec en Outaouais for their insightful contributions.

Seminar Summary

With the growing use of artificial intelligence across various domains — including the Internet of Things (IoT), healthcare, and transportation — it has been demonstrated that AI systems, particularly those based on deep learning, are vulnerable to adversarial attacks capable of deceiving them.

Although numerous defense methods have been proposed to mitigate this threat, these same methods can also be attacked when they are known to adversaries. In such cases, attackers can design adversarial examples that bypass defenses and successfully compromise AI solutions.

The main challenge, therefore, is not only to develop methods capable of defending against adversarial attacks in artificial intelligence and machine learning, but also to ensure that these defense mechanisms themselves are robust and secure.

In this presentation, we provide an overview of existing approaches to mitigating adversarial attacks and highlight their limitations. We also present our state-of-the-art methods for defending against adversarial attacks in AI through secure defense mechanisms.