A Hierarchical Two-tier Approach to Hyper-parameter Optimization in Reinforcement Learning

TL;DR

This paper introduces a hierarchical Bayesian optimization method for tuning hyper-parameters in reinforcement learning, improving automation and potentially enhancing RL performance in control tasks.

Contribution

It proposes a novel two-tier Bayesian optimization approach that separately optimizes categorical and solution-level hyper-parameters in RL.

Findings

Promising results in a simulated control task.

Effective separation of hyper-parameter optimization levels.

Potential for more user-independent RL applications.

Abstract

Optimization of hyper-parameters in reinforcement learning (RL) algorithms is a key task, because they determine how the agent will learn its policy by interacting with its environment, and thus what data is gathered. In this work, an approach that uses Bayesian optimization to perform a two-step optimization is proposed: first, categorical RL structure hyper-parameters are taken as binary variables and optimized with an acquisition function tailored for such variables. Then, at a lower level of abstraction, solution-level hyper-parameters are optimized by resorting to the expected improvement acquisition function, while using the best categorical hyper-parameters found in the optimization at the upper-level of abstraction. This two-tier approach is validated in a simulated control task. Results obtained are promising and open the way for more user-independent applications of reinforcement learning.