What is Reinforcement Learning?

Reinforcement Learning

• x → y
  • x: reward or the reward function
  • positive reward / negative reward

Applications

• Controlling robots
• Factory optimization
• Financial (stock) trading
• Playing games (including video games)

Mars rover example

The return in reinforcement learning

• Return = $R_1+r{R}_2+r^2{R}_3+...$ (until terminal state)
• Discount Factor: $r=0.9$
  • $0+(0.9)0+(0.9{)}^{2}0+(0.9{)}^{3}100$
• The return depends on the actions you take

Making decisions: Policies in reinforcement learning

Policy

• A policy is a function mapping from states to actions, that tells you what action to take in a given state
  • $\pi (s)=a$
  • $\pi$ : policy
  • $s$ : state
  • $a$ : action

The goal of reinforcement learning

• Find a policy $\pi$ that tells you what action $(a=\pi (s))$ to take in every state $s$ so as to maximize the return

Review of key concepts

Markov Decision Process (MDP)