I completed my PhD within the tau team under the direction of Sylvain Chevallier and Guillaume Charpiat, and during which I worked on the expressivity of neural networks. In particular, I have proposed and implemented
a Neural Architecture Search strategy which jointly optimizes a network architecture and its weights using a new metric named the Expressivity Bottleneck. This metric associates a network architecture's location with its lack of expressivity by quantifying the network's ability to follow its functional gradient.
I started a post-doctoral position in the Ockham team under the direction of Rémi Gribonval in december 2026 and I am currently studying changes of Hilbert spaces in neural networks optimization. Indeed, each Hilbert space is associated with a scalar product, which itself defines a gradient that is an optimization direction for the network parameters.
Through my academic courses, I have been studying statistics and classical machine learning tools such as Linear Regression, Random Forest, SVM, and their constrained variants. Since my PhD, I changed my object of study and took an interest in neural networks and the understanding of their behaviors when solving one problem or another.
My research domain is at the crossroads of geometry, optimization, and statistics.
