Coursebooks

Theory and Methods for Reinforcement Learning

EE-618

Lecturer(s) :

Cevher Volkan

Language:

English

Frequency

Every 2 years

Remarque

Every 2 years. Next time: Spring 2021

Summary

This course describes theory and methods for decision making under uncertainty under partial feedback.

Content

1. Introduction to the reinforcement learning (RL) paradigm

2. Overview of classical developments I: Markov Decision Process (MDP, POMDP), and Dynamic Programming (Value Iteration, Policy Iteration)

3. Overview of classical developments II: Monte-Carlo methods, TD-Learning, Q-Learning, SARSA (Model-based RL, and Model-free RL)

4. Stochastic Bandits and Thompson (posterior) Sampling

5. Bandit based RL algorithms (UCRL, UCAgg, UCCRL, REGAL) - Exploration and Exploitation

6. Policy Search (Policy gradient algorithms, variance reduction, TRPO algorithm)

7. Imitation Learning (Inverse Reinforcement Learning, Apprenticeship Learning)

Keywords

Reinforcement learning, policy search.

Learning Prerequisites

Required courses

Optimization, probability theory, mathematics of data.

Assessment methods

Project report.

In the programs

    • Semester
    • Exam form
       Project report
    • Credits
      3
    • Subject examined
      Theory and Methods for Reinforcement Learning
    • Number of places
      20
    • Lecture
      28 Hour(s)
    • Practical work
      14 Hour(s)

Reference week

 
      Lecture
      Exercise, TP
      Project, other

legend

  • Autumn semester
  • Winter sessions
  • Spring semester
  • Summer sessions
  • Lecture in French
  • Lecture in English
  • Lecture in German