Research on the performance of grid computing for distributed neural networks

In this paper, we report the first evaluation of cooperation computing for artificial neural networks in distributed environment. Several performance-relevant factors are considered, including architecture of computing service, workflow and cooperation strategy. Evidence on basic processes and performance of such strategies of cooperation computing are reviewed. We also present a theoretical analysis of distributed-training strategies of neural networks for structure-distributed and data-distributed. We prove a strategy of distributed computing based on data-distributed is more feasibility for distributed neural networks, which makes training the neural networks more efficient. In the final, we concluded the evaluation by briefly considering selected open questions and emerging directions in construction of grid computing for distributed neural networks.