Reinforcement Learning for Playing Othello

Eker T.A. (2017) Reinforcement Learning for Playing Othello. Bachelor's Thesis, Artificial Intelligence.

Preview	Text AI-BAI-2017-T.A.Eker.pdf - Published Version Download (277kB) | Preview
	Text Toestemming.pdf - Other Restricted to Backend only Download (80kB)

Abstract

In this thesis we examined the effect of adding more hidden layers, using different activation functions and using a dropout algorithm on the performance of an artificial neural network (ANN) learning to play Othello using Q-learning. While all of these methods have seen promising results in supervised learning, the results for our reinforcement learning experiment were not that spectacular. The activation function with the best results is the sigmoid function. The ReLU and ELU functions do not increase the performance of the agent nor do they speed up the convergence of the ANN. Dropout also does not increase performance. Adding more hidden layers increased the performance and helped speed up the learning process. An ANN with two hidden layers using a sigmoid function and no dropout algorithm yields the best performance in our experiment for playing Othello and wins 95% of the testing games playing against a fixed opponent after learning for two million games.

Item Type:	Thesis (Bachelor's Thesis)
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	15 Feb 2018 08:29
Last Modified:	15 Feb 2018 08:29
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/15445

Actions (login required)

View Item