- Understand the Reinforcement Learning problem and how it differs from Supervised Learning
- Reinforcement Learning (RL) is concerned with goal-directed learning and decision-making.
- In RL an agent learns from experiences it gains by interacting with the environment. In Supervised Learning we cannot affect the environment.
- In RL rewards are often delayed in time and the agent tries to maximize a long-term goal. For example, one may need to make seemingly suboptimal moves to reach a winning position in a game.
- An agent interacts with the environment via states, actions and rewards.
Lectures & Readings
- Reinforcement Learning: An Introduction - Chapter 1: The Reinforcement Learning Problem
- David Silver's RL Course Lecture 1 - Introduction to Reinforcement Learning (video, slides)
- OpenAI Gym Tutorial