Investigation into the effect of social learning in reinforcement learning board game playing agents

摘要

This thesis presents the use of social learning to improve the performance of gameudplaying reinforcement learning agents. Agents are placed in a social learning environmentudas opposed to the Self-Play learning environment. Their performance is monitored andudanalysed in order to observe how the performance changes compared to Self-Play agents.udTwo case studies were conducted, one with the game Tic-Tac-Toe and the other with theudAfrican board game of Morabaraba. The Tic-Tac-Toe agents used a table based TD ( )udalgorithm to learn the Q values. The results from the tests for the Tic-Tac-Toe agentsudindicate that the social learning agents perform better than the Self-Play agents in bothudboard tests and competitive tests. By increasing the population sizes of the agents theudnumber of superior social agents also increases as well as improvements in their skilludlevel. In the second case study the agents use function approximation and the TD ( )udalgorithm because of a larger number of states. The social agents performed better thanudthe Self-Play agents in the board tests and are not superior in the test where they competeudagainst each other. Larger populations were not possible with the Morabaraba agents butudthe results are still positive as the agents perform well in the board tests.