Simon Roburin

I am a PhD student at Imagine team at Ecole des Ponts Paris Tech (ENPC) and under the supervision of Mathieu Aubry, Renaud Marlet and Patrick Perez.

CV  /  Google Scholar

profile photo

I'm interested in computer vision, machine learning and optimization. My research is mainly about understanding deep neural network with a specific focus on their robustness.

Spherical Perspective on Learning with Batch Norm
Simon Roburin*, Yann de Mont-Marin*, Andrei Bursuc, Renaud Marlet, Patrick Perez Mathieu Aubry
arXiv, 2020
project page / arXiv / code

Leveraging radial invariance in CNN with Batch Norm (BN) to build a spherical framework which provides interesting theoretical insights of how BN interacts with optimization. Introduction of a variation of Adam, AdamSRT which improves signifiantly performances over a variety of datasets and architectures.