Projects.

Understand. Build.
Put it to the test.

Explore projects

01 07AI learning platform

DeepLearn

Learn to play better, and advance your AI.

Classes adapted to your level, a workshop to build and train your networks, then an arena to put them to the test. The duels and the Elo ranking give a concrete objective to learning.

Discover the project

02 07Classification of exoplanet candidates

Exoplanet Hunter

Analyze observations from Kepler, K2 and TESS.

An application for ranking exoplanet candidates and reviewing their predictions. The project opens a thesis axis: developing a Transformer from scratch combining light curves, physical data and TSFRESH statistics.

Discover the project

03 07Deep Reinforcement Learning and tree search

Reinforcement gaming agents

Train and compare agents, grids and Quarto.

Implementation of DQN, Double DQN and REINFORCE, compared to move search methods. The presentation details the models, the results and their limitations, then allows you to observe them in play or to confront them.

Discover the project

04 07Unsupervised learning

Understanding latent space

Understand how models represent and organize images.

Exploring the latent space: a representation where a few coordinates summarize an image. With PCA, autoencoders and VAEs, the experiments and workshops show how this space is built, what it preserves and what happens when we move there.

Discover the project

05 07Design of an industrializable RAG

Documentary and statistical assistant

Link documents, statistics and geography in natural language.

A wizard for non-sensitive data, designed for enterprise deployment. It combines large models, documentary research, calculations and maps by limiting the information transmitted to the API. The technical choices and methods of verifying the results are detailed in the project presentation.

Discover the project

06 07Tabular methods and planning

Reinforcement learning

Compare algorithms on nine game environments.

Study of Q-learning, SARSA, Monte Carlo and Dyna-Q, with a reference to dynamic programming. From grids to Monty Hall, experiments allow you to compare exploration, rewards and strategies, then train an agent.

Discover the project

07 07Implementation from scratch

Machine learning library

Develop linear models, MLPs and RBF networks.

A Rust library accessible from Python, with training, prediction and saving of models. The presentation explains calculations and backpropagation; a workshop allows you to compare decision boundaries on 2D data.

Discover the project