Adaptive Implementation of Artificial Neural Networks Reflecting Changing Hardware Resources at Run-Time.
Udo Seiffert · 2005
A number of basic mathematical algorithms along with their ability of a hardware adaptive implementation actually already paved the way towards more general and also more complex frameworks. However, artificial neural networks, which often become rather complex and which are inherently suitable and – at least when applied to largescale data sets – often required to be run on parallel hardware, have not moved into the focus of real hardware adaptive implementations yet. This paper systematically reviews the state-of-the-art and the requirements of implementing artificial neural networks on varying parallel computer hardware. Based on this it provides perspectives which clearly extend recent attempts of hardware adaptive implementations based on generally varying but at run-time fixed resources.