Deep Reinforcement Learning

3 Credits · Level: Graduate, Undergraduate

Deep Reinforcement learning is an area of machine learning that learns how to make optimal decisions from interacting with an environment. From the environment, an agent observes the consequence of its action and alters its behavior to maximize the amount of rewards received in the long term. Reinforcement learning has developed strong mathematical foundations and impressive applications in diverse disciplines such as psychology, control theory, artificial intelligence, and neuroscience. An example is the winning of AlphaGo, developed using Monte Carlo tree search and deep neural networks, over world-class human Go players. The overall problem of learning from interaction to achieve goals is still far from being solved, but our understanding of it has improved significantly. In this course, we study fundamentals, algorithms, and applications in deep reinforcement learning. Topics include Markov Decision Processes, Multi-armed Bandits, Monte Carlo Methods, Temporal Difference Learning, Function Approximation, Deep Neural Networks, Actor-Critic, Deep Q-Learning, Policy Gradient Methods, and connections to Psychology and to Neuroscience. The course has lectures, mathematical and programming assignments, and exams.