Large Neural Network Simulations on Multiple Hardware Platforms
Efficient simulations of very large networks of interconnected neurons require particular consideration of the computer architecture being used. Techniques for implementing simulators on a number of different computer architectures are presented.
The experience gained from adapting an existing simulator, SWIM, to two very different architectures, vector computers and multiprocessor workstations, is analyzed. This work led to the implementation of a new simulation library, SPLIT, designed to allow efficient simulation of large networks on several different architectures. SPLIT hides from the user most of the architecture dependent details, that is, the particular data structures and computational organization actually utilized.
KeywordsComputer Architecture Efficient Simulation Vector Computer Computational Organization Lamprey Spinal Cord
Unable to display preview. Download preview PDF.
- Orjan Ekeberg, Per Hammarlund, Björn“Levin, and Anders Lansner. SWIM — A simulation environment for realistic neural network modeling. In Josef Skrzypek, editor, Neural Network Simulation Environments. Kluwer, Hingham, MA, 1994.Google Scholar
- Per Hammarlund and Örjan Ekeberg. Large neural network simulations on multiple hardware platforms, 1996. (Submitted.).Google Scholar
- [31.Anders Lansner and Erik Fransén. Improving the realism of attractor models by using cortical columns as functional units. In James M. Bower, editor, The Neurobiology of Computation: Proceedings of the Third Annual Computation and Neural Systems Conference, pages 251–256, Monterey, CA, July 21–26, 1994 1995. Kluwer, Boston, MA.Google Scholar
- William W. Lytton. Optimizing syaptic conductance calculation for network simulations. neucomp,8(3):501509, 1996.Google Scholar
- Tom Wadden, Jeanette Hellgren-Kotaleski, Anders Lansner, and Sten Grillner. Intersegmental coordination in the lamprey — simulations using a continuous network model, 1996. In Press.Google Scholar