MIT: Deep Reinforcement Learning

Explore deep reinforcement learning in this comprehensive lecture from MIT's Introduction to Deep Learning course. Delve into key concepts of reinforcement learning, including Q-functions, policy gradients, and their applications. Learn about different classes of learning problems, how to define and utilize Q-functions for action selection, and the intricacies of deep reinforcement learning algorithms. Discover the downsides of Q-learning and the advantages of policy gradient methods. Examine real-world applications, including the groundbreaking achievements in the game of Go with AlphaGo and AlphaZero. Gain insights into the latest advancements in AI and machine learning through this informative presentation by Alexander Amini.

Intro
Classes of Learning Problems
Reinforcement Learning (RL): Key Concepts
Defining the Q-function
How to take actions given a Q-function?
Deep Reinforcement Learning Algorithms
Digging deeper into the Q-function
Downsides of Q-learning
Policy Gradient (PG): Key Idea
Policy Gradient (PG): Training
The Game of Go
AlphaGo Beats Top Human Player at Go (2016)
AlphaZero: RL from Self-Play (2018)