A problem solving environment that assists model development for reinforcement learning algorithms

Abstract

This paper reports a problem solving environments (PSE) to assist researchers who study stochastic simulations such as reinforcement learning algorithms. They have to run their programs many times to compare their algorithms and find better sets of parameters for their programs. In order to reduce the working time, this system has three sub-systems: a distributed computing system, a data management system and a graph generation system. Using this system, we conduct experiments with human subjects. They register their programs, run them on a distributed computing system, obtain results automatically, and compare them graphically. As a result, a user obtained five times speedup for the work time. We present a relationship between development of algorithms and the three sub-systems.