Computational Complexity Reduction of Secure BP Learning Algorithm with Divided Data

Abstract

Many studies have been conducted on how to perform learning while maintaining data security. One of them is the secure distributed processing with decomposition data. The feature of this method is that it achieves high confidentiality by decomposing data and parameters in machine learning and performing distributed processing. On the other hand, realization of machine learning by distributed and integrated processing of decomposed data and parameters leads to an increase in computational complexity and degradation of computational accuracy. In this paper, we propose an improved BP learning method that suppresses the increase in computational complexity with the increase in the number of servers, and demonstrate its effectiveness.