Learn what an action was worth

A journey through reinforcement learning with Botie.

Adapting and Evaluating Modern Models · A journey with Botie

First visit? Learn the workshop controls

Botie's notebook

How Do We Learn When the Consequence Arrives Later?

Part 1 of 3

Suppose an agent is moving through a maze. We may know when it reaches the exit without having a labelled correct move for every location. Some useful moves initially take it away from the exit, so judging each action only by its immediate reward can be misleading.

Sources for this history

Watkins and Dayan (1992), Q-Learning

← Previous lesson: Low-Rank Adaptation Read the full lesson and derivations