Nathan Rousselot

Thales DIS, Meyreuil, France

Hello and welcome!

I’m currently pursuing a PhD as part of the SmartIES team at LIRMM, hosted within the Advanced Security Team at Thales Silicon Security. My research is guided by an incredible supervisory team: Loïc Masure (LIRMM), Karine Heydemann (Thales DIS), Vincent Migairou (Thales DIS), and Philippe Maurine (LIRMM). My primary focus is on deep learning-based side-channel analysis, with a strong emphasis on its theoretical foundations and challenges.

I hold a Master degree in Statistical Machine Learning for Information Processing at ENSEEIHT, part of the French Grandes Écoles. During my studies, I explored applied mathematics at KU Leuven in their "Mathematical Engineering" master's program as an exchange student and studied Computer Engineering at Efrei Paris.

Research Interests

1. Side-Channel Analysis and Deep Learning:
   I’m fascinated by the unique challenges of side-channel analysis, particularly its formulation as a deep learning problem. My current focus is on understanding the Plateau Effect (Masure et al., 2023) and finding strategies to overcome this limitation. I’m also exploring connections between information theory and deep learning-based side-channel analysis.

2. Optimization in Deep Learning:
   I have a keen interest in second-order optimization algorithms and Bregman-based descent methods, which have yet to find appropriate use cases in deep learning.

Previous Research Experiences

1. IRIT Student Researcher:

   • (June to December 2021 & June to August 2022)
     I conducted statistical processing of photoplethysmography signals under the supervision of Henrique Goulart (IRIT-SC) and Marie Chabert (IRIT-SC). Later, I collaborated with Wissal Ghamour (INP-N7) to approximate nonlinear (linear-analytic) physical models using Volterra Kernel-based methods, also supervised by Henrique Goulart.

2. R&D Intern at Exail Sonar Division:

   • (June to September 2023)
     During my time at Exail, I worked on large-scale statistical analysis of multi-beam echo sounder data, under the supervision of Guillaume Matte (Exail) and Marie Lamouret (Exail).