Reinforcement Learning (RL) is a Machine Learning (ML) paradigm that focuses on goal-directed learning from interactions. In particular, it is learning how to map situations to actions by maximising a scalar reward signal. RL is studied in other disciplines such as game theory, robotics, operations research, and multi-agent systems. It has roots in psychology and the advancements in psychology contributed to the advancements of RL and vise versa. It has been existing since some years, has been highlighted and gained again the attention of ML researchers in the recent years, specially in terms of Deep Reinforcement Learning (DRL). In this course, we will cover the difference between RL and other ML paradigms such as supervised learning and unsupervised learning, exploration and exploitation dilemma, main elements of RL systems, model-free, and model-based methods, planning, control and how to design RL algorithms to RL problems.
At the end of the course you will:
- know the fundamentals of reinforcement learning
- know different RL problems and solutions
- implement RL methods
|Week 1||26.04.2020||Lecture||Bohlouli||Introduction to the course and logistics||Course_Intro.pdf|
|28.04.2020||Lecture||Bohlouli||An Introduction to Reinforcement Learning||RL_Intro.pdf|
|Week 2||03.05.2020||Tutorial||Bohlouli||Markov Decision Process||MDP.pdf|
|05.05.2020||Tutorial||Nazeri||Elements of RL Systems: Environments||Blackboard|
|Week 3||11.05.2020||Lecture||Bohlouli||Planning by Dynamic Programming||Planning_DP.pdf|
|12.05.2020||Lecture||Bohlouli||Components of RL Agent||RL_Comp.pdf|
|19.05.2020||Tutorial||Nazeri||Elements of RL Systems: Action Values||Blackboard|
|Week 5||24.05.2020||Lecture||Bohlouli||Public Holiday|
|Assignment #||Release Date||Description||Submission Deadline||Source Files|
|Assignment 1||05.05.2020||Soccer environment||19.05.2020, 16:59 CEST||soccer_env.zip|
|Assignment 2||19.05.2020||Action Values||01.06.2020, 23:59 CEST||Assignment 2, Supp_materials|
|Assignment 3||02.06.2020||15.06.2020, 23:59 CEST|
|Assignment 4||16.06.2020||29.06.2020, 23:59 CEST|
|Assignment 5||30.07.2020||13.07.2020, 23:59 CEST|
Course Student Presentations:
|12.05.2020||Policy Gradient Methods for Reinforcement Learning with Function Approximation|
|19.05.2020||A Natural Policy Gradient||Amirreza Mohammadi|
|26.05.2020||Approximately Optimal Approximate Reinforcement Learning||Ehsan Rassekh|
|02.06.2020||An Analysis of Temporal-Difference Learning with Function Approximation|
|09.06.2020||Reinforcement Learning of Motor Skills with Policy Gradients|
|16.06.2020||Algorithms for inverse reinforcement learning|
|23.06.2020||Robot learning from demonstrations||Sarina Danaei|
|30.06.2020||The Dynamics of Reinforcement Learning in Cooperative Multiagent Systems||Fatemeh Merhara|
|07.07.2020||Hysteretic Q-Learning : an algorithm for decentralized reinforcement learning in cooperative multi-agent teams||Reza Khaleghi|
|14.07.2020||Kernel-Based Reinforcement Learning|
- have experiences and good knowledge of machine learning
- be familiar with linear Algebra
- have solid programming skills in Python
- be familiar with working on a Unix-style operating systems
Reinforcement Learning (Second Edition), by Richard S. Sutton and Andrew G. Barto, Stanford University, 2018.
Deep Reinforcement Learning Hands-On (2nd Edition), by: Maxim Lapan, Packt Publishing; 2 edition (31 Jan 2020).
Class Time and Location:
- Sundays, 14:30 – 16:00 CET.
- Tuesdays, 14:30 – 16:00 CET.
- Given the current Corona Situation, this semester, the class will be completely online.
- You should first apply for approval through the following link. This link will be also the online course sessions every week.
- Course Videos Link: [https://vuniv.iasbs.ac.ir/b/dr–xea-944]
- Final exam will be in the written form.
- The final exam will be held on 27.06.2020, at 06:30 CET.