Seminar: State Representation for Deep Learning in Large State and Action Spaces

Rory Campbell
M.Sc. Thesis Proposal
Supervisor: Dr. David Churchill

State Representation for Deep Learning in Large State and Action Spaces

Department of Computer Science
Friday, November 23, 2018, 1:00 p.m., Room EN 2022


Abstract

Artificial intelligence (AI) is an integral part of both game development and play. AI that is dynamic
enough to provide a level of challenge appropriate to a variety of players can improve a game’s reception for both audiences and critics. Strategy games are often complex due to the large number of possible scenarios they present. Given how challenging strategy games are, the AI players in these games have been historically weak. Solutions to these complex problems have utility for AI in general, extending beyond the games in which they are implemented.

Machine Learning (ML) is a sub-field of AI which encompasses many analytical and learning techniques useful in writing a sophisticated AI game-player. As a testing platform for my work, I will be working with the Canadian video game company Lunarch Studios on incorporating new ML techniques into the existing AI framework of their retail video game, Prismata. By improving the AI system within Prismata, it will strengthen their product and provide their users with an overall better gaming experience. Due to the complex nature of the game, any technique capable of performing well in this environment should be applicable to other industries as well.