Courses Detail Information

ECE6705J – Deep Reinforcement Learning

Instructors:

Credits: 3 credits

Pre-requisites: Necessary computer science and mathematical background covered in VE281, VE203, VV216/256/286

Description:

Reinforcement learning (RL) is a generic artificial intelligence approach for autonomously learning from trials and error. Deep RL is the integration of deep learning and reinforcement learning. This course will cover the main recent algorithms and approaches in deep RL and illustrate them on various applications (traffic light control, video games, robotics…).
This course provides a presentation of deep RL, with a focus on its recent developments. Topics include RL basics, deep Q-learning, policy gradients, actor-critic algorithms, model-based RL, imitation learning, inverse RL, hierarchical RL, and multi-agent RL.

Course Topics:

Introduction to reinforcement learning
Varkov decision processes and variants
Planning
Overview ofmachine learning
Behavior cloning
Model-free policy evaluation
Model-free policy learning
Value function approximation
Policy gradient
Actor-critic methods
Vidterm review
Advanced value-based deep RL methods
Vidterm exam
Advanced policy-based deep Rl methodsPitfalls in deep RL
Off-policy learning and entropy regularized learning
Integrating model learning and policy learning
Learning from demonstration and human preferences
Exploration-exploitation dilemma
Multi-objective reinforcement learning
Multi-agent reinforcement learning
Other topics (if time permits): Offine RL, safe RL, or muiti-task RL