This course is about reinforcement learning (RL), covering the fundamental concepts of reinforcement learning framework and solution methods. The focus is on the underlying methodology as well as theory. The course will cover the following topics, including foundations of RL, dynamic programming, Monte Carlo methods, temporal difference learning and Q-learning.

This short course is made based on materials in Sutton and Barto (2018), Puterman (1994), DeepMind & UCL RL lecture series and some research papers.  It covers the foundations of reinforcement learning, planning and learning.

The course website is available at https://github.com/callmespring/RL-short-course.

主讲人：

Chengchun Shi，Associate Professor in data science at London School of Economics and Political Science

时间：8月17日至8月19日，9:00-12:00，共3次课

地点：浙江大学数据科学研究中心，行政楼1417

本次短期课程计划从以下方面讨论强化学习：

第一部分：Foundations of Reinforcement Learning

1. Introduction to Reinforcement Learning

1.1 Multi-armed Bandits

1.2 Contextual Bandits

2. Markov Decision Processes

2.1 Time-varying Markov Decision Processes

2.2 Partially Observable Markov Decision Processes

3. Existence of the Optimal Stationary Policy

第二部分：Planning and Learning

1. Preliminaries

2. Planning: Dynamic Programming

2.1 Policy Iteration

2.2 Value Iteration

3. Learning: Monte Carlo Methods

3.1 Monte Carlo Prediction (Policy Evaluation)

3.2 Monte Carlo Control (Policy Optimization)

4. Learning: Temporal Difference Methods

4.1 Temporal Difference Prediction

4.2 Temporal Difference Control: SARSA

第三部分：Q-Learning and Beyond

1. Tabular Q-learning

2. Fitted Q-iteration

3. Case Study 1: Deep Q-networks

4. Case Study 2: Reinforcement Learning for Ridesharing