非模块型人工神经网络并行实现的研究

STUDY ON PARALLEL IMPLEMENTATION OF NON MODULAR ARTIFICIAL NEURAL NETWORK

本文对两种非模块型神经网络——全连接和随机连接神经网络在多处理器网络上并行实现的性能进行了分析 ,指出处理器网络的拓扑结构和结点的扇入对这两种神经网络的并行实现的性能影响不大 ,并对神经网络在多处理器网络上并行实现时一个学习周期时间进行了分解 。

The performance of parallel implementation for two kinds of non modular artificial neural networks—fully and random connected artificial neural networks on multiprocessor architecture is analyzed in this paper. It concludes that the topology of the multiprocessor architecture and the fan in of processor have little effective. The iteration learning time of neural networks is discussed, and the maximum speedup and best size of multiprocessor system are calculated in this paper too.