A Theory of Abstraction in Reinforcement Learning

TL;DR

This paper develops a theoretical framework for abstraction in reinforcement learning, proposing criteria for effective abstraction functions and introducing algorithms to learn such abstractions efficiently, ultimately reducing complexity in decision-making.

Contribution

It introduces a formal theory of abstraction in reinforcement learning, with new algorithms and analysis to learn abstractions that preserve behavior, are efficient, and reduce planning complexity.

Findings

Proposed three desiderata for abstraction functions.

Developed algorithms to learn abstractions meeting these criteria.

Provided analysis showing reduced complexity in reinforcement learning.

Abstract

Reinforcement learning defines the problem facing agents that learn to make good decisions through action and observation alone. To be effective problem solvers, such agents must efficiently explore vast worlds, assign credit from delayed feedback, and generalize to new experiences, all while making use of limited data, computational resources, and perceptual bandwidth. Abstraction is essential to all of these endeavors. Through abstraction, agents can form concise models of their environment that support the many practices required of a rational, adaptive decision maker. In this dissertation, I present a theory of abstraction in reinforcement learning. I first offer three desiderata for functions that carry out the process of abstraction: they should 1) preserve representation of near-optimal behavior, 2) be learned and constructed efficiently, and 3) lower planning or learning time. I then present a suite of new algorithms and analysis that clarify how agents can learn to abstract according to these desiderata. Collectively, these results provide a partial path toward the discovery and use of abstraction that minimizes the complexity of effective reinforcement learning.