Zacharie Rodière

Currently a research intern in the Physics Laboratory at ENS de Lyon, France. Working with Pierre Borgnat and Paulo Gonçalves, I apply machine learning techniques to neuroscience, specifically focusing on detecting Seizure Onset Zones (SOZ) from stereo-EEG signals in epilepsy patients. My recent work resulted in an accepted extended abstract at the Graph Signal Processing Workshop (MILA, Montréal, May 2025).

I'm also a Master's student in Electrical and Computer Engineering at Georgia Tech, where I'll be starting my Master's thesis in Fall 2025.

Email / CV / LinkedIn

Research

I am interested in foundational representation learning, with a focus on effective training methodologies and pre-training strategies. My work primarily explores contrastive learning frameworks, including supervised variants that leverage label information. A key application area for these techniques has been the use of Transformer architectures for stereo-EEG signal processing. I also have experience adapting Transformer models for computer vision tasks.
Previously, I have applied related representation learning concepts using Graph Neural Networks, though my current focus is more aligned with the broader paradigms underlying representation learning. Looking ahead, I am keen to explore the potential of these techniques within Natural Language Processing and Reinforcement Learning.

Leveraging clinically-informed time-frequency features and spatial contrastive pre-training within a Transformer encoder for improved Seizure Onset Zone (SOZ) localization from stereo-EEG.

Developed and evaluated a Vision Transformer-based system for multi-label classification of defects on historical book bindings to aid conservation efforts.

Website design from Jon Barron. Favicon graphic by Collaborapix Studio on Vecteezy.com.