Platforms for artificial neural networks : neurosimulators and performance prediction of MIMD-parallel systems
L.G. Vuurpijl · Radboud Repository (Radboud University) · 1998
In this thesis, two platforms for simulating artificial neural networks are discussed: MIMD-parallel processor systems as an execution platform and neurosimulators as a research and development platform. Because of the parallelism encountered in neural networks, distributed processor systems seem to provide a proper underlying execution platform. The suitability of the class of MIMD-parallel computer platforms (in particular multi-transputer systems) for neural network simulation programs is discussed in this thesis. In order to evaluate the suitability of such systems, a new performance prediction method is presented and evaluated for several classes of neural networks. Neurosimulators provide a platform for simulating, developing, evaluating and executing neural network models. In the last two chapters of this thesis, neurosimulators are examined: environments for the development and simulation of artificial neural networks. By considering their common features, and the requirements of their users, the design criteria for a new neurosimulator are specified. The design, implementation and evaluation of PREENS, an action-oriented neurosimulator is presented in the final chapter.