This site serves as my personal repository for sharing scientific and ML deep-dives and insights.
Currently, I work on the predictive side of a hedge fund, where I architect and deploy various deep learning models, including large language models (LLMs) and Transformers, for predictive tasks across diverse data modalities. My interests span text, audio, and video data, as well as complex graph structures.
My background includes a double MSc degree in Engineering School (ENSAE Paris) and Management (ESSEC Business School), followed by a Ph.D. in Applied Maths / Machine Learning completed at Intitut Polytechnique de Paris (IPP) and Criteo. My doctoral research specialized in adversarial attacks targeting both Computer Vision models and rankingss used in recommender systems.
PhD Thesis - Statistical Understanding of Adversarial Robustness
I have successfully defended my PhD thesis on the 30th of November 2023!
(Paper) AISTATS 2023 - Origins of Low-Dimensional Adversarial Perturbations
I am thrilled to announced that my paper called Origins of Low-Dimensional Adversarial Perturbations with Elvis Dohmatob and Chuan Guo from Meta AI Research has been accepted and will be published at AISTATS 2023! It has been selected to be one of the 32 papers to have an “oral” presentation :) Go check it for more details.
(Insights) Focus on Cross-entropy loss and KL divergence
Deep learning model massively use the classical Cross-Entropy loss for training. Let’s look at this loss in more details
(Paper) MLSS and DLRL 2020 Poster Video
As part of the MLSS and DLRL 2020 virtual organization, I recorded a “poster video” about a paper we - with Thomas Ricatte and Elvis Dohmatob from Criteo AI Lab - recently submitted to NeurIPS 2020.
(Researcher's Corner) Work Organization for PhD Student
Like many other PhD student before, I have been facing some challenging issues regarding organization of my work since the beginning of my PhD. I have asked advice from collegues or other PhD students, and researched on the internet any tip that could be useful.
(Researcher's Corner) ArXiv Alert
ArXiv is an incredible tool to get access to research papers, but it has one drawback for me: there are too much papers in it! I want to filter new ArXiv based on my research interests, so here is how I created an ArXiv Alert.
(Insights) Of Rabbits and Foxes
A quick overview of the Lotka-Volterra model.