IMPLEMENTATION OF PARALLELIZING MULTI-LAYER NEURAL NETWORKS BASED ON CLOUD COMPUTING

Abstract

Background: Cloud computing, as a technology developed under the rapid development of modern network, is mainly used for processing large-scale data.  The traditional data mining algorithms such as neural network algorithm are usually used for processing small-scale data.  Therefore, the calculation of large-scale data using neural network algorithm must be based on cloud computing.  Materials and Methods: Firstly a Hadoop system was established taking MapReduce as the programming framework.  Then the parallelized traditional data mining algorihtm was investigated based on cloud computing cluster to verify its feasibility in processing large-scale data.  Finally, the speed and training precision of the algorithm were tested. Results: It was feasible to process large-scale data with cloud computing based parallelizing multi-layer neural network algorithm. The speed of parrallel processing was faster if data size was larger, especially if the sample was in a size of more than 1 million.  It was more superior to the serial back propagation network in training preciseness.  Conclusion: Parallelizing multi-layer neural network based on cloud computing platform can process large-scale data effectively in the perspectives of time and quality.