Reinforcement Learning Performance Evaluation: An Evolutionary Approach

Reinforcement Learning Performance Evaluation: An Evolutionary Approach

0

PRE

AI

DOI：

原文链接

原文求助

摘要

En 中文

In this paper, we analyze the effect of metaparameters on the performance of actor-critic reinforcement learning. The considered two different applications: first the pole-balancing benchmark task and then a surviving behavior where the Cyber Rodent robot has to capture the battery packs and increase its own energy level. The results show that metaparameters highly influence the learning time. In addition, optimal metaparameters generated by evolutionary algorithms have mutual relation with each other.

Keyword:

Reinforcement learning

metaparameters

evolutionary algorithm

Reinforcement Learning Performance Evaluation: An Evolutionary Approach

摘要

期刊

机构

引用论文