Playing Ms. Pac-Man using Advanced Neural Networks

Caat, A.J. ten (2017) Playing Ms. Pac-Man using Advanced Neural Networks. Bachelor's Thesis, Artificial Intelligence.

Preview	Text AI_BA_2017_AJTENCAAT.pdf - Published Version Download (619kB) | Preview
	Text Toestemming.pdf - Other Restricted to Backend only Download (78kB)

Abstract

In this research, we have used several types of neural networks, differing in the amount of hidden layers and if they use a sigmoid function or ReLU to calculate the activations. These networks, along with Q-learning, have been used to play Ms. Pac-man. Previous research already used the same approach, but with one hidden layer and only sigmoid activation functions, to show the effectiveness of higher-order action-relative inputs. The same inputs were used here, along with an extra addition. The final results of the networks from this research were compared with those of the other research, both when the environment remained the same during training and testing, as well as when the environment differs. The outcome was that the additions made to the original network did indeed lead to a significantly better performance. The networks were also compared to each other, which showed that ReLU activation networks perform significantly better than sigmoid activation networks when the environment stays the same from training to testing, and that there is an interaction between the activation function used and the amount of hidden layers when the environment differs.

Item Type:	Thesis (Bachelor's Thesis)
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	15 Feb 2018 08:26
Last Modified:	15 Feb 2018 08:26
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/14915

Actions (login required)

View Item