This notebook provides a brief introduction to reinforcement learning, eventually ending with an exercise to train a deep reinforcement learning agent with the dopamine framework.

The notebook is roughly organized as follows:

- A simple motivating example is presented to illustrate some of the main points and challenges.
- Markov decision processes (MDPs), the mathematical formalism used to express these problems, is introduced.
- Exact tabular methods are presented, which can be used when the environment is fully known. These methods form the foundation for the learning algorithms in uncertain environments.
- Value-based learning algorithms are introduced, and related algorithms are also presented.
- The methods introduced are put to the test on a simple MDP.
- Deep reinforcement learning is introduced, along with some implementation details.
- We use deep reinforcement learning to learn a policy on a larger MDP.
- Resources for further study are provided.

Don't forget to tag @psc-g in your comment, otherwise they may not be notified.

señor swesearcher at Google Brain. Musician.

Reinforcement Learning Tutorial

2020-07-02 ·
Important reinforcement learning (RL) algorithms, including policy iteration, Q-Learning, and Neural Fitted Q.

Short Notes on Batch Constrained Deep Reinforcement Learning

2020-06-12 ·
Blog article on Off-Policy Deep Reinforcement Learning without Exploration paper by Fujimoto et al. (ICML 2019)

Acme: A Research Framework for Reinforcement Learning

2020-06-01 ·
A library of reinforcement learning components and agents.

Short Notes on Model-Based Offline Reinforcement Learning (MOReL)

Blog article on Model-Based Offline Reinforcement Learning (MOReL) paper by Rahul Kidambi & Aravind Rajeswaran et al.