人工神经网络在并行计算机集群上的设计研究

ON DESIGNING ARTIFICIAL NEURAL NETWORKS ON PARALLEL COMPUTER CLUSTER

人工神经网络在集群上的并行化设计和实现能够充分发挥ANN并行处理的特点,缩短训练时间,降低算法复杂度。随着并行技术的日益成熟,在并行集群上以软硬件相结合的方式设计神经网络的重要性也不断提高。从软硬件平台的多方面讨论了并行集群技术对人工神经网络设计的支持,提出了一种SOM神经网络在并行集群上的设计方法和基础框架,并就并行集群上神经网络训练效率的问题进行了深入讨论。该方案可广泛应用于多种神经网络模型的并行计算机实现。

Parallel design and implementation of artificial neural networks on clusters can give full play to the characteristics of ANN in parallel processing,shorten the training time and reduce the algorithm complexity.With parallel technology having become more matured,the neural networks＇ design process of hardware and software combination is getting more importance in these days.The support of parallel cluster technique on artificial neural network designing is discussed in terms of various aspects of both hardware and software platform.A new design method and framework of SOM neural network on parallel cluster is presented,and the efficiency of neural network training on parallel cluster is discussed in-depth.The artificial neural network parallel design can be widely used in a variety of neural network model on parallel computer.