OPTIMIZING THE SOM FOR THE PLAYSTATION 2

Abstract – The use of neural networks in computer game AI is a rapidly expanding area of academic research; however, the games industry has been slow to adopt new techniques, and one reason behind this is the scarcity of computational resources in modern games. The Sony PlayStation 2 ® provides parallel hardware features that can be exploited to implement a highly optimized SOM. This paper summarises the main hardware features of the platform, and presents some optimizations that can be made to the SOM algorithm to tailor it to the hardware. This work makes the SOM a viable option for AI in games, where processing resources are used intensively. The principles presented are also applicable to other modern parallel processors such as “The Cell, ” and is therefore relevant to the next generation of games consoles and other devices. Key words – optimization, games, parallel processin