Fast and Secure Back-Propagation Learning Using Vertically Partitioned Data with IoT

Abstract

The use of the cloud computing system is expanding. As the number of terminals (things) connected to the cloud system increases, the limit of the capability is also becoming apparent. It leads to significant processing time delay. Edge (or fog) computing system is known as a method for improving a conventional cloud system. The basic idea is to consider a system that places edges (servers) between the cloud and the terminals (things). The capacity of each edge may not be so high, but many edges cooperate to execute tasks to achieve high processing power. Then, how should machine learning be realized on the edge system? Fast and secure learning methods are desired for machine learning. The use of a cryptographic system does not seem to be necessarily suitable for machine learning. Therefore, a safe system using distributed processing has attracted attention. SMC (Secure Multiparty Computation) is one of the typical models of them. Horizontally and vertically partitioned data are known for SMC. There have been proposed some methods for realizing machine learning on the cloud using SMC. Also, some methods of machine learning using horizontally partitioned data of SMC on the edge system have been proposed. On the other hand, little studies have been done on machine learning using the vertically partitioned data. In this paper, a fast and secure BP (Back-Propagation) neural network learning on vertically partitioned data with edge system is proposed. The effectiveness is shown by numerical simulation.