Large Scale Biologically Realistic Models of Cortical Microcircuit Dynamics with Application to Novel Statistical Classifiers (Pilot Investigation)

摘要

The purpose of this project was to better understand brain-like network dynamics by incorporated biological parameters into large-scale computer simulations using parallel distributed 'Beowulf' clustering. Milestones included improved single- processor efficiency of 24-fold. On multiprocessor clusters, initial time trials on - networks of 2 to 1000 cells suggests that the total time does not depend as heavily on the product (connection probability) x (N cells)squared, but shows a substantial * linear term. The projected time to run a 1-million cell simulation would be about 5.6 days on a single CPU, or roughly 6 hours on the proposed 30-CPU Beowulf system. Substantial progress was made toward a C++ implementation for subsequent research.