IRL — Inverse Reinforcement Learning

Fact

A method for guessing the hidden reward rule from expert actions.

In Plain Words

IRL is like copying Grandma’s cookies with no recipe. You watch her hands and guess her secret scorecard.

It works backward from examples to learn preferences for robots, self-driving cars, and game AI.

Related Concepts

• Reinforcement Learning
• Imitation Learning
• Markov Decision Process

RL
IRL learns the reward first, then RL finds the best plan.

Imitation Learning
IRL also watches expert demos, but it cares about the hidden goal.

MDP
IRL often uses an MDP to define states, actions, and rewards.