Artificial neural networks on massively parallel computer hardware

摘要

The implementation of artificial neural networks (ANNs) on powerful parallel computer hardware is closely related to the simulation of ANNs on general purpose computers itself. Although there are many different good reasons for a parallel implementation, there has always been a lot of scepticism as well. The neural networks community seemed to be divided into those who do and those who do not. However, along with the continuously increasing availability of powerful and reasonably priced computer clusters within the last few years, this topic appears in a new light, which should be a reason enough to generally review it. This paper gives a survey of the state-of-the-art parallel computer hardware from a neural networks user's point of view and guides those people who are willing to go the way of a parallel implementation utilising the most recent and accessible parallel computer hardware and software. In order to emphasise the tutorial character of this paper, it is rounded off with an extensive reference section.