QVA-learning: Testing a Novel Reinforcement Learning Algorithm Using Other-Play in the Helenix Environment

Heeres, K.J. (2020) QVA-learning: Testing a Novel Reinforcement Learning Algorithm Using Other-Play in the Helenix Environment. Bachelor's Thesis, Artificial Intelligence.

Preview	Text AI_BA_2020_JobHeeres.pdf Download (3MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (96kB)

Abstract

In this paper we use the Helenix environment and other-play, to test a new reinforcement learning algorithm called QVA-learning. This algorithm builds upon QV-learning by adding a function which should improve the learning behaviour. Within the game of Helenix, 5 different algorithms will train against each other using other-play. We found that when comparing the final models, Q-learning performed best under these conditions and that QVA-learning managed to outperform both double Q-learning and its predecessor QV-learning. However, when looking at the entire learning process, only SARSA manages to outperform QVA-learning. We conclude that QVA-learning shows potential and improves upon its predecessor QV-learning.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Wiering, M.A.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	02 Sep 2020 07:15
Last Modified:	02 Sep 2020 07:15
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/23344

Actions (login required)

View Item