About

Research engineer specialized in statistics and deep learning applied to computer vision and time series modelling. PhD from the Institut Polytechnique de Paris. Experienced with several programming languages for both research and production. Research interests include visual search engines, generative models and uncertainty estimation.

Co-founder of deepnet, an association aiming at solving real life problems by using artificial intelligence. We are helping students skill up in the filed of AI, connect them with like minded individuals, and deploy solutions to social and economic issues.

Projects

FarmAi

The FarmAi project aims at using our knowledge in data science to better understand how to grow plants. Currently, our work revolves around using the FarmBot to track various information on growing plants, such as the room and soil temperature, humidity, salinity, as well as a visuals of the crops. We hope to release a consistent dataset soonTM.

Transformers for time series

Transformers are cool, but back in the days, not easily applied to time series data, for reasons ranging from its quadratic complexity in the time dimension to the attention mechanism that can get out of hand on very long sequences, to the nature of the originally proposed positional encoding. We proposed an adapted implementation from scratch in the repo Transformer for Time Series. This work is best described in our paper End-to-end deep meta modelling to calibrate and optimize energy consumption and comfort.

Librairie Le Phare

I have developed and still maintain the website for the Librairie Le Phare, available at librairielephare.fr.

Publications

Max Cohen, Sylvain Le Corff, Maurice Charbit, Alain Champagne, Marius Preda and Gilles Nozières (2023). "End-to-end deep meta modelling to calibrate and optimize energy consumption and comfort." Energy and Buildings (Hal).

Max Cohen, Guillaume Quispe, Sylvain Le Corff, Charles Ollion and Eric Moulines (2022). "Diffusion bridges vector quantized Variational AutoEncoders." International Conference on Machine Learning (ICML) (Hal, Slides and presentation).

Max Cohen (2023) "Métamodèles et approches bayésiennes pour les systèmes dynamiques." Thèses.fr.

Max Cohen, Maurice Charbit and Sylvain Le Corff (2023). "Variational Latent Discrete Representation for Time Series Modelling." Accepted in the IEEE Workshop on Statistical Signal Processing (SSP) 1 (Hal).

Max Cohen, Maurice Charbit and Sylvain Le Corff (2023). "Last layer state space model for representation learning and uncertainty quantification." Accepted in the IEEE Workshop on Statistical Signal Processing (SSP) 1 (Hal).

1This paper was accepted, but required one of the authors to physically attend the conference to be published. In order to avoid a 12 hour long flight (and back!), we decided to decline the invitation.