Research on Automated Reinforcement Learning: based on Tree-structured Parzen Estimators and Median Pruning
Zhaolei Wang, Ludi Wang, Qi Liang, Wuyi Luo, Qinghai Gong, Shanshan Li · 2021
Tuning the hyper parameters of machine learning algorithm is regarded as a boring and difficult challenge to the researcher in the artificial intelligence domain. However, with the rapidly development of computer clusters and GPU processors, the automated machine learning algorithm have been proposed to solve this problem. In this paper, an automated reinforcement learning method has been proposed by focusing on the automated hyper parameters tuning of the reinforcement learning, a relatively new branch of machine learning which is more suitable to the motion control. For reducing the amount of calculations and finding the optimization solution more quickly, an open automated optimization framework have been proposed by combing the tree-structured Parzen estimators based Bayesian optimization and median pruning algorithm. A specific simulation has been given based on the deep deterministic policy gradient case to show the effectiveness and practicability of the proposed method.