2 An Overview

 

The intent of this section is to appeal to the intuitions of the reader, leaving much of the detail to later sections in the paper. The subsections that follow will demonstrate in turn: that there are features in the function produced by reinforcement learning; that graphs based on these features can be used to control the composition of the function pieces; that these features are easy to detect early in the learning process; that these features exist in multiple domains.

• 2.1 Features in the Reinforcement Learning Function
• 2.2 Composing Function Pieces
• 2.3 Detecting Features Early
• 2.4 A Different Task Domain