MIT: Reinforcement Learning

This lecture introduces deep reinforcement learning for learning in dynamic environments without fixed labeled data. It covers Q-functions, deep Q networks, policy gradients, discrete and continuous action spaces, and applications including games and autonomous driving.

Syllabus

Intro
Learning in Dynamic Environments
Classes of Learning Problems
Reinforcement Learning (RL): Key Concepts
Defining the Q-function
Deep Reinforcement Learning Algorithms
Digging deeper into the Q-function
Deep Q Network Summary
Downsides of Q-learning
Discrete vs Continuous Action Spaces
Policy Gradient (PG): Key Idea
Training Policy Gradients: Case Study
Reinforcement Learning in Real Life
Reinforcement Learning and the Game of Go
Deep Reinforcement Learning Summary