Reinforcement Learning

This course was written by Tessa van der Heiden, a researcher and developer of autonomous driving algorithms at BMW.

In this course, you'll learn the mathematical underpinnings of reinforcement learning, a foundational machine learning technique in which an agent (or algorithm) is trained by trial and error. By rewarding the agent for good outcomes, it "learns" optimal strategies, which can be applied to problems in domains like robotics, quantitative trading, and game theory.
This course is intended for young professionals who are interested in applying machine learning techniques for decision making, or students who are pursuing a machine learning career or preparing for interviews.

• Introduction:
• Introduction: How does a computer devise a strategy to play a game optimally?
• Foundations:
• Value Functions: When transitioning between various options, the algorithm must quantify how good these options are.
• Dynamic Programming: Optimize an interconnected system by reducing it into smaller systems.
• Monte Carlo: If we make random moves a large number of times, we might notice a pattern that allows us to solve the problem deterministically.
• Extensions:
• Temporal Difference Learning: Explore a method of reinforcement learning that updates every time step — not just at the end of the episode.
• Policy Gradient Methods: These methods take a different approach — by learning the optimal policy directly.