Interplay between AI and cybersecurity: robustness and explainability of machine learning models

Ronan Hamon
European Commission, Joint Research Centre & LIS, QARMA, Aix-Marseille Université
https://www.researchgate.net/scientific-contributions/Ronan-Hamon-2052785845

Date(s) : 28/05/2021   iCal
14 h 30 min - 15 h 30 min

The increased uptake of Artificial Intelligence (AI) technologies in industry and society leads to a stronger reliance on digital systems, with higher potential impacts in case of cybersecurity incidents or infringements on fundamental rights. In particular, the use of machine learning techniques brings a new class of vulnerabilities that pose new kinds of technical challenges. In this presentation, I will focus on two specific challenges: First, the challenge of explainability, linked to the opaqueness of machine learning models, will be discussed through a comparison between technical explanations and legal requirements as set out in the General Data Protection Regulation. Second, the challenge of adversarial robustness will be introduced through a case study on autonomous driving, describing in particular how adversarial machine learning techniques can be leveraged to attack and deceive classification and detection models.

Catégories

• Séminaire Signal et Apprentissage