CEC 2013 Fiesta Americana Grand Coral Beach Hotel Grand Coral Ballroom Chichen Itza and Tulum Isla Contoy Isla Mujeres Ruinas del Rey Cozumel and Hel-Ha Kayaking and Windsurfing Tres rios and Actun Chen


Computational Intelligence and Games

Simon M. Lucas, Philip Hingston and Julian Togelius


In recent years, the field of Computational Intelligence and Games (CIG) has enjoyed rapid progress and a sharp rise in popularity. In this field, algorithms from across the computational intelligence spectrum are tested on benchmarks based on e.g. board games and video games, and new CI-based solutions are developed for problems in game development and game design. This tutorial will give an overview of key research challenges and methods of choice in CIG. The tutorial is divided in two parts, where the second part builds on methods and results introduced in the first part. The tutorial will include videos of spectacular examples of CI in games, live demonstrations and concrete tips on how to get your own CIG-related research started.

Part 1: Learning to play games

Learning to play games is arguably the core challenge in CIG research. A variety of learning algorithms have been applied to games as diverse as Chess, Go, car racing, Pac-Man and Unreal Tournament. We discuss how evolutionary algorithms and temporal difference learning can be used to learn game strategies and generate NPC behaviours, and give several examples from the literature. A key design choice that strongly affects the performance of the learning algorithm is how to represent the strategies: we introduce and compare several strategy representations, including neural networks, fuzzy logic, interpolated tables and n-tuple classifiers. We also discuss emerging techniques for CI-based game-playing, such as Monte Carlo tree search.

Part 2: Player modelling and procedural content generation

There are many uses for computational intelligence in games besides learning how to play the game optimally. This part of the tutorial begins by revisiting the reasons for using CI in games, and then proceed to survey two particularly important areas: modelling player behaviour and preferences, and creating game content automatically. In order to create human-like characters in games and tailor games to the preferences of particular players, we need to be able to model both the behaviour and the preferences of real human players. An opportunity here is the vast amounts of telemetry data automatically collected by game developers, e.g. via the XBox Live network. We present examples of extracting player types and predicting player behaviour using machine learning techniques. One use of modelling players is to better adapt games to individual players, for example by procedurally generating game content (such as levels, items and scenarios) that suits particular players. Procedural content generation can also be used to cut costs and increase creativity in the game development process, and enable new kinds of games. We discuss different approaches to procedural content generation, focusing on the search-based approach that builds on evolutionary algorithms.


Biography Philip Hingston

Philip Hingston is a Senior Member of IEEE and is Chair of the IEEE Computational Intelligence Society's Technical Committee on Games. He runs the annual BotPrize competition, and recently edited a new book on game AI/CI: Believable Bots. He is the author of more that 90 conference papers and journal articles. He holds the B.Sc. degree in pure mathematics from the University of Western Australia, and the Ph.D. degree in mathematical logic from Monash University. Currently, he is an Associate Professor in the School of Computer and Security Science, Edith Cowan University, Australia.


The length of the tutorial:
four hours.


Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer