I am a postdoctoral researcher in deep learning and computer vision at EPFL in the Visual Intelligence for Transportation (VITA) lab, advised by Prof. Alexandre Alahi. My research objective lies in advancing the state of AI by building learning systems that are robust, reliable and aligned with shared human values. In particular, I am currently investigating how to leverage the recent advances of LLMs in multimodal learning for autonomous driving applications.

Before joining EPFL, I completed my PhD thesis on robust deep learning for autonomous driving in March 2022 at Conservatoire National des Arts et Métiers and valeo.ai. I was fortunate to benefit from the supervision of Prof. Nicolas Thome and Dr. Patrick Pérez. Prior to my PhD, I worked as a computer vision engineer at Earthcube and as a research intern at Heuritech. I graduated from Master 2 Data Sciences at Ecole Polytechnique in 2017 and I obtained a master’s degree in computer science from Ecole Centrale de Lille in 2016.

UPDATES

• 2023.05: A pleasure to discover the Transport research community at the 23rd Swiss Transport Research Conference where I also presented our on-going work on pedestrian action recognition with visual-language models.

• 2022.11: Joined the Visual Intelligence for Transportation (VITA) lab at EPFL as a postdoctoral researcher. Looking forward to applying deep learning to computer vision for autonomous driving and robotics, under the supervision of Prof. Alexandre Alahi.

• 2022.03: Successfully defended my PhD thesis, Robust Deep Learning for Autonomous Driving, and I’m now officially a doctor!

• 2021.07: I presented our paper Beyond First-Order Uncertainty Estimation with Evidential Models for Open-World Recognition at the ICML Workshop on Uncertainty and Robustness in Deep Learning. In this work, we tackle the challenge of detecting model misclassifications and out-of-distribution samples in a single measure.

• 2021.05: Our paper Confidence Estimation via Auxiliary Models has been accepted in IEEE T-PAMI journal! Among other things, we use our learned confidence approach ConfidNet to improve domain adaptation for semantic segmentation.

• 2021.01: I attended the Mediterranean Machine Learning Summer School from January 11th to January 16th. Such great lectures and interesting people!

• 2020.12: Our new paper on using auxiliary models to better estimate confidence of neural networks available here. We extend ConfidNet’s approach to self-training in unsupervised domain adaptation for semantic segmentation interested.

• 2020.11: We have a 6-month research internship to work on an exciting project about Dirichlet networks and out-of-distribution detection. Details available here. If interested, contact me or my PhD advisor, Nicolas Thome.

• 2019.10: Talk at the GdR ISIS (CNRS) seminar Théorie du deep learning in October 17th about uncertainty and failure prediction with deep neural networks [slides]

• 2019.10: Code for ConfidNet is now available at https://github.com/valeoai/ConfidNet .

• 2019.09: Our paper Addressing Failure Prediction by Learning Model Confidence has been accepted at NeurIPS 2019;

• 2019.01: Started my PhD on Robust and Multimodal Deep Learning for Autonomous Driving in January 2019 at CEDRIC (CNAM) and valeo.ai, under the supervision of Nicolas Thome and Patrick Pérez.

• 2017.10: I presented our paper Leveraging Weakly Annotated Data for Fashion Image Retrieval and LabelPrediction written during my research internship at Heuritech at Computer Vision for Fashion Worskhop at ICCV 2017.