This course focuses on the Deep Reinforcement Learning. Deep
Reinforcement Learning (DRL) is a part of Artificial Intelligent (AI). In DRL system, it is composed of agent, environment, and reward function. The system is encouraged to take actions that lead to positive reward functions. This common pattern is the foundation of deep reinforcement learning: building machine learning systems that explore and learn based on the responses of the environment.

The students on the completion of this course would be able to:
• Apply knowledge learned to build an intelligent system using Deep Reinforcement Learning
• Use appropriated DRL algorithms to solve problems such as control problems, gaming playing etc.
• Design and develop a software-based system with DRL integration

None

Module 1: Classical Reinforcement Learning
I. Introduction to Deep Learning
1. Past, preset, and future of deep learning
2. Mathematical foundation
3. Applications
II. Exploration vs. Exploitation
1. The challenge of interpreting evaluative feedback
2. Exploration vs. Exploitation
III. Algorithms
1. Bellman Ford algorithms
2. Dynamic Programming
3. Temporal Differences
4. Monte Carlo
IV. Model based learning
1. Finite Markov Decision Process
Module 2: Deep Reinforcement Learning
V. Value based learning
1. SARSA
2. Q Network
3. Dyna Q Network
4. Deep-Q Network
5. Double Q Learning
VI. Policy based learning
1. Policy gradient methods
2. VPG learning
Module 3: Advanced Deep Reinforcement Learning
VII. Actor critic based learning
1. A2C model
2. DDPG model
VIII. Future reinforcement learning
1. Parallel reinforcement learning
2. Toward AGI: Artificial General Intelligence

Basic Reinforcement Algorithms, Artificial Neural Network, Deep
Reinforcement Learning, Actor Critic based Learning

1. R. S. Suttun, and A. G. Barto: Reinforcement Learning, MIT Press, 2nd edition, 2018

1. A. Zai, B. Brown: Deep Reinforcement Learning, Manning, 1st edition, 2020
2. M. Morales: Grokking Deep Reinforcement Learning, Prentice Hall, 1st edition,2020
3. L. Graesser, W. L. Keng, Foundation of Deep Reinforcement Learning, Addison- Wesley, 1st edition, 2020

IEEE Intelligent Systems, IEEE
IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE
IEEE Transactions on Neural Networks and Learning Systems, IEEE
IEEE Transactions on Knowledge and Data Engineering, IEEE
Pattern Recognition, Elsevier
Neurocomputing, Elsevier

Lectures: 30 hours
Laboratory sessions: 45 hours
Presentations: 3 hours
Self-study: 90 hours

Lecture with hand-on lab. There will be project at the end of semester when students conduct research project.

The Final grade will be computed according to the following
components:

midsemester exam 20%;

final exam 20%;

assignments and laboratories 10%
and project 50%.

Open-book examination is used for both mid-term and final exam.
An “A” would be awarded if a student can demonstrate clear understanding of the knowledge learned in class as well as from the laboratory assignments and literature reviews.
A “B” would be awarded if a student can understand the basic principles of the knowledge learned in class, from the laboratory assignments and from literature reviews.
A “C” would be given if a student can understand partially the basic principles of the knowledge learned in class, from the laboratory assignments and from literature reviews.
A “D” would be given if a student shows lack of understanding of the knowledge learned in class, from the laboratory assignments and from literature reviews.