Profile Image

I am a research scientist at Meta AI, working in AI for neural interfaces.

Prior to joining Meta, I worked in deep learning for automotive applications at Valeo.ai, with a particular emphasis on computer vision and human pose estimation.

Before my time at Valeo, I was a postdoctoral researcher at Inria’s MIND team (previously known as Parietal team). Under the supervision of Thomas Moreau and Alexandre Gramfort, I delved into the intersection of deep learning and neuroscience. My primary research focused on automatic data augmentation, specifically learning optimal data augmentations directly from datasets. I was particularly interested in applications for which data augmentations are not intuitive, such as EEG (electroencephalography), where data augmentations served as a tool to better understand how certain types of information, like sleep stages, are encoded within neurological signals. Over time, my work transitioned into the field of geometric deep learning, with a focus on learning and exploiting data symmetries to enhance the efficiency and robustness of neural networks.

Before my postdoctoral position, I served as the scientific and engineering leader of the AI team at Ava. Our primary focus was on speaker recognition technology for deaf and hard-of-hearing accessibility. My team and I were particularly focused on real-time speaker diarization, i.e., the difficult task of automatically detecting “who said what when”.

I obtained my PhD in applied mathematics from Ecole Polytechnique (CMAP) and INRIA, under the supervision of Frédéric Bonnans and Pierre Martinon. My research centered on the intersection of optimal control, machine learning, and optimization. I focused on developing interpretable and physically plausible models of dynamical systems, and on optimally controlling them while accounting for model uncertainty. My thesis was funded by the aviation startup Safety Line (later acquired by SITA), and the primary application of my work was the optimization of real aircraft trajectories for fuel consumption reduction. The algorithms I developed were integrated into the product OptiClimb, which is currently used daily to compute fuel-efficient flights globally by companies such as AirFrance.

Prior to my PhD, I obtained a MSc. in engineering and applied mathematics from MINES ParisTech.

Current research interests

(non-exhaustive)

Reviewer

Teaching

rss facebook twitter github youtube mail spotify lastfm instagram linkedin google google-plus pinterest medium vimeo stackoverflow reddit quora quora