Portrait of Romain Laroche

Romain Laroche



I am part of the Microsoft Research Montreal Reinforcement Learning team.

My research interest focuses on Reinforcement Learning. It consists in learning through trial and error to control an agent behaviour in a stochastic environment: at each time step, the agent performs an action, and then perceives from its environment an observation, and receives a reward. My preferred application domains are dialogue systems (real-world motivation), Atari games (for benchmarking against other algorithms), and navigation toy problems (for empirical analysis and algorithm design). Lately, I focus more specifically on reliability of RL algorithms.