About

I am currently a postdoctoral fellow in the Statistics Department of ENSAE-CREST, working with Anna Korba. I received my PhD in Mathematical Computer Science at Inria, where I was advised by Aapo Hyvärinen and Alexandre Gramfort.

My research is in machine learning. More specifically, I work on estimating and sampling energy-based models, on density-ratio estimation, and on representation learning for brain imaging data. My latest publication is on optimal distribution paths for annealed importance sampling.

You can reach me at emir.chehab [AT] ensae.fr

Publications

Self-Supervised Learning and Statistical Estimation

Provable benefits of annealing for estimating normalizing constants: Importance Sampling, Noise-Contrastive Estimation, and beyond
O. Chehab, A. Hyvärinen, A. Risteski
Spotlight, Neural Information Processing Systems (NeurIPS), 2023.
Paper Poster Code

Annealing with the right paths exponentially increases the sample-efficiency of some estimators.

Optimizing the Noise in Self-Supervised Learning: from Importance Sampling to Noise-Contrastive Estimation
O. Chehab, A. Gramfort, A. Hyvärinen
Submitted, 2023.
Paper Code

Unnormalized distributions are identified by their energy and normalizing constant. Importance sampling estimates the latter, while Noise-Contrastive Estimation estimates both: we draw a formal connection.

The Optimal Noise in Noise-Contrastive Learning Is Not What You Think
O. Chehab, A. Gramfort, A. Hyvärinen
Uncertainty in Artificial Intelligence (UAI), 2022.
Paper Poster Code

In many setups, the noise distribution that minimizes the estimation error is very different from the data distribution.

Deep Learning and Cognitive Neuroscience

Deep Recurrent Encoder: A scalable end-to-end network to model brain signals
O. Chehab*, A. Defossez*, J.C. Loiseau, A. Gramfort, J.R. King
Journal of Neurons, Behavior, Data analysis, and Theory, 2022.
Paper

A deep learning architecture best predicts brain activity caused by visual stimuli. It uses the interaction between the initial cognitive state and the visual stimulus for prediction.

Uncovering the structure of clinical EEG signals with self-supervised learning
H. Banville, O. Chehab, A. Hyvärinen, D. Engemann, A. Gramfort
Journal of Neural Engineering, 2021.
Paper

Self-supervised learning can produce features that correlate with sleep stages, pathology, and other neurophysiological markers.

A mean-field approach to the dynamics of networks of complex neurons, from nonlinear Integrate-and-Fire to Hodgkin-Huxley models
M. Carlu*, O. Chehab*, [...], A. Destexhe, M. di Volo
Journal of Neurophysiology, 2020.
Paper

Mean-Field analysis effectively summarizes complex network dynamics that model neuronal activity.

Teaching

I was Teacher’s Assistant for the following Masters courses.

Optimization for Data Science - Institut Polytechnique de Paris (2021-2023)
Professors: Alexandre Gramfort, Pierre Ablin

Advanced Machine Learning - CentraleSupelec, Universite Paris-Saclay (2020-2022)
Professors: Emilie Chouzenoux, Frederic Pascal

Optimization - CentraleSupelec, Universite Paris-Saclay (2020-2021)
Professors: Jean-Christophe Pesquet, Sorin Olaru, Stephane Font