基于MapReduce的RBF神经网络的并行性研究

Research on Parallelism of RBF Neural Network Based on MapReduce

针对串行传统的单机训练学习算法很难满足大数据应用需求,本文讨论了基于云平台计算机集群的神经网络并行化问题。首先,设计实验验证基于云平台机群的神经网络并行化学习的可行性;其次,将此方法应用到人脸识别;最后,通过仿真实验验证方法的有效性,与非并行化径向基网络进行对比,表明该方法可以降低数据训练时长,有更快的学习效率、更高的识别度。

It is difficult for the traditional serial single training learning algorithm to meet the de- mand of large data application, thus this paper discusses the parallel computer cluster neural network based on the cloud platform. First of all, to design experiment based on the cloud platform to validate the feasibility of the neural network parallel learning; Secondly, the method is applied to face recogni- tion; Finally, through the simulation results to verify the validity of the method, compared with the parallel radial basis networks. Data show that the method can reduce the training time, can have faster learning efficiency and higher degree of recognition.