Position: Benchmarking is Limited in Reinforcement Learning Research

TL;DR

This paper examines the limitations of benchmarking in reinforcement learning research, highlighting the high computational costs and proposing alternative experimental paradigms to improve evaluation practices.

Contribution

It identifies the computational challenges of rigorous benchmarking and advocates for supplementary evaluation methods to address these limitations.

Findings

Benchmarking in RL is computationally expensive.

Current practices often lead to misleading claims.

Alternative evaluation paradigms are needed.

Abstract

Novel reinforcement learning algorithms, or improvements on existing ones, are commonly justified by evaluating their performance on benchmark environments and are compared to an ever-changing set of standard algorithms. However, despite numerous calls for improvements, experimental practices continue to produce misleading or unsupported claims. One reason for the ongoing substandard practices is that conducting rigorous benchmarking experiments requires substantial computational time. This work investigates the sources of increased computation costs in rigorous experiment designs. We show that conducting rigorous performance benchmarks will likely have computational costs that are often prohibitive. As a result, we argue for using an additional experimentation paradigm to overcome the limitations of benchmarking.