Communication Load Reduction for Neural Network Implementations on Message Passing Multicomputers

Abstract

Simulation speed of neural network implementations on message passing multicomput-ers can be enhanced by a reduction of the total communication load, which could be achieved using near-optimal partitioning of neural networks. We consider in particular the impact of both network structure and sparseness on the communication load. Our simulation results… (More)

Topics

  • Presentations referencing similar topics