Reinforcement Learning: An Introduction

Pezzulo, Rigoli and Friston (2015) cite this book as the standard reference for reinforcement learning. In their account, prediction errors, as in Temporal Difference learning, serve mainly as teaching signals, and responses are triggered by stimuli through associative links. They contrast this with Active Inference, where goals trigger actions through prediction errors, which they place closer to cybernetic regulation. They also note that Active Inference likewise uses prediction error as a teaching signal, but with the learning rate mediated by precision.

actions are triggered by goals through prediction errors, whereas in most reinforcement learning schemes, responses are triggered by stimuli via associative links

Pezzulo G, Rigoli F, Friston K, 2015 · Active Inference, homeostatic regulation and adaptive behavioural… · open at passage 63

prediction error is used as a teaching signal for learning – as in standard reinforcement learning models

Pezzulo G, Rigoli F, Friston K, 2015 · Active Inference, homeostatic regulation and adaptive behavioural… · open at passage 51