/images/toff.png

AI Research Engineer @ Inria

Hi I'm Corentin, a Research Engineer in the Flowers team. My interests include Machine Learning, Reinforcement Learning, Evolutionary Strategies, and LLMs. I'm currently seeking a position in one of these areas.

Research: Evolving Reservoirs for Meta Reinforcement Learning

Paper presented at EvoStar 2024 (Long Talk). Abstract Animals often demonstrate a remarkable ability to adapt to their environments during their lifetime. They do so partly due to the evolution of morphological and neural structures. These structures capture features of environments shared between generations to bias and speed up lifetime learning. In this work, we propose a computational model for studying a mechanism that can enable such a process. We adopt a computational framework based on meta reinforcement learning as a model of the interplay between evolution and development.

Hackathon: Ebiose

Introduction Participated to a project called Ebiose during a 2 days hackathon with a team of 4 people. It focuses on a multi-language-based-agent system to solve tasks. In this system, agents are composed of LLM (e.g. GPT-3.5) coupled with a specific prompt. The structure of the multi-agent system followes a predetermined sequence, where one agent answers a question, another verifies the answer, and a final agent synthesizes both responses. This collaboration is inspired by the Autogen Framework, emphasizing coordinated efforts among agents.

Software: Rising Sun

To try to beat friends on the combat phases of the “Rising Sun” board game, I created this simplified gymnasium implementation of the game. The project also incorporates a python script to train a single RL agent on the environment. This agent can currently only be trained against hard-coded players (either random of with heuristics). The next step will be to add multi-agents training to the project. Then, it will be interesting to explore the path of self play to see how complex the behaviors of the agent can become !

Software: Genetic Algorithm for sport's pools optimization

As a volleyball player who frequently travels to play against teams in distant cities, I became curious about how sports pools are optimized to ensure fair and reasonable travel times for teams in the same category. To explore this further and delve into evolutionary algorithms, I embarked on a project to create a custom genetic algorithm using NumPy. The custom genetic algorithm is designed to improve the current distribution of cities in sports pools.

Research: Symbolic Reinforcement Learning

Paper published in HAL-Inria (Inria’s Research or Technical Reports) during a semester project in my last MSc year. Abstract Modeling complex problems is a real issue in Reinforcement learning. The goal of this projet is to use a RL model, such as Q-learning, on symbolic data, where the knowledge of the problem would be stored, in the form of qualitative or numerical data (in JSON for example). Instead of only working with numerical data, we could link the data with a certain ’edition distance’ which permits to evaluate the proximity of 2 elements.

Alzheimer disease detection by 3D MRI analysis

This project focuses on medical imaging and computer vision. The goal is to develop a machine learning model able to detect the state of development of the Alzheimer disease, using 3D MRI data from real patients. In order to do so, we tried different approaches, including the ‘2D epsilon’ method, which is the use of 2D Convolutionnal neural networks on 2D slices of the MRI, and also the use of 3D CNN.