Machine Learning is the study of how software agents learn to make good predictions or decisions from experience or data.

Today: AI / Data-driven decision making. → How can software agents learn to make good decisions from experience or data?

Main branches of machine learning:

• Supervised learning: predictions from labeled data;

  

• Unsupervised learning: predictions from labeled data;

  

• Reinforcement learning: optimal decisions from data and experience

  

Reinforcement Learning

Goal: Learn to make optimal sequences of decisions in an unknown environment by trial and error.

Basic setting

Agent:

• Receives an observation of the current state
• Selects an action
• Receives a reward from the environment

Environment:

• Receives an action from the agent
• Gives a reward to the agent
• Changes the environment state

This setup might look somewhat similar to human learning.

RL vs. other ML tasks