The success of deep learning came with several issues: vulnerabilities against adversarial attacks, concern regarding privacy leakages of the train set, problematic behavior for out-of-distribution data, instable training in challenging tasks with high variance gradients, and difficulties to reason properly in LLM. The overarching objective of my research is to develop learning algorithms that are more robust to these uncertainties and defaults.
In particular, I focused on optimal transport and robustness against adversarial attack, through their common denominator: Lipschitz constrained neural networks. I also studied privacy preserving machine learning.
My current interest goes toward ``optimization as a layer’’ paradigm, computer graphics and algorithmic information theory.
An example of research statement encompassing my recent interests and summarizing my contributions can be found here.
PhD thesis and defense
The first draft of my thesis can be found here.
The defense will take place February 7th, 2024 at 9am in the IRIT Auditorium.