Dynamic Programming and Reinforcement Learning

URL study guide

https://studiegids.vu.nl/en/courses/2024-2025/XM_0093

Course Objective

Dynamic Programming (DP) and Reinforcement Learning (RL) are fields concerned with decision making over time. After completing this course, the student
- is familiar with the commonly used algorithms for solving dynamic optimization problems;
- understands the main features of these algorithms, their strengths and weaknesses including their convergence properties;
- can implement them in an appropriate language;
- can model real-world decision problems into a DP or RL framework and solve moderately sized problems;
- has knowledge of the historical development of DP and RL and has an idea of possible future developments.

Course Content

This course is concerned with reinforcement learning and its origin dynamic programming. These are fields dealing with goal-directed decision making over time, such as finding your way in an unknown area, playing a game or pricing airline tickets. We look at these areas from different angles:
- we deal with full-information "planning" problems, but also with partial-information "learning" problems
- we consider different algorithms, some of which are guaranteed to find the best solution, but also heuristics
- we consider high-dimensional problems (such as games) and methods to solve them
- we look at small toy problems to understand algorithms and sharpen our intuition, but also bigger problems for which we learn how to implement algorithms (in python)
- we look at different types of applications, both from AI (search problems, games) and OR

Teaching Methods

Lectures and practical work integrated

Method of Assessment

Programming exercises and final exam. The 3 assignments each count for 10% and the exam for 70%. The minimal passing grade for the exam is 5.0.

Literature

Slides and lecture notes

Target Audience

mBA, mAI, mCS, mBa-D, mMath

Recommended background knowledge

Programming experience in python