CentraleSupélec — Reinforcement Learning

Reinforcement Learning — CentraleSupélec

A rework of the PARL course, structured as four lectures, each followed by a written-exercise sheet picking up the topic of the lecture before it.

LaTeX sources live in a private repository; the distributed PDFs are published to github.com/cetiennec/cours-pdf and served straight from GitHub — click a title below to preview it inline, or “open” for the full-screen PDF.

Lectures

1 — Introduction and MDPs open ↗

2 — Dynamic Programming open ↗

3 — Model-Free Prediction open ↗

4 — Model-Free Control & Deep RL Applications open ↗

Session 1: Markov decision processes, return, value functions, Bellman equations. Session 2: policy evaluation, policy iteration, value iteration. Session 3: Monte-Carlo methods, temporal differences TD(0). Session 4: Monte-Carlo control, SARSA, Q-learning, and a look at deep RL.

Written exercises

Exercises — Introduction and MDPs open ↗

Exercises — Dynamic Programming open ↗

Exercise sheets for sessions 3 and 4, and the notebook tutorials, are in preparation.

See also the Deep Reinforcement Learning training I run for companies — same fundamentals, geared towards hands-on practice with Stable-Baselines3 and Gym.