Summary
International Symposium on Nonlinear Theory and its Applications
2017
Session Number:B1L-B-1
Session:
Number:B1L-B-1-3
Parallelization of a Spiking Neural Network Model of Layered Cortical Sheet Consisting of Multiple Cortical Regions
Jun Igarashi,
pp.375-378
Publication Date:2017/12/4
Online ISSN:2188-5079
DOI:10.34385/proc.29.B1L-B-1-3
PDF download (187.1KB)
Summary:
A parallel computing of spiking neural networks of the cortex at whole-brain scale is a grand challenging in the next decade. In a whole-brain scale simulation, load imbalance and increasing communication of spikes reduce computational efficiency. To overcome the problems, we investigated tile partitioning parallelization of a spiking neural network model of the cortex with layer structure using supercomputer K. We added one communication feature to reduce communication frequency using spike propagation delay of long-range connections. The parallelization showed reduction of communication frequency and elapsed time, and ideal scaling performance. The tile partitioning parallelization may work for a simulation of the cortex at whole-brain scale.