Society 5.0 aims to integrate cyber space with real space. In these days, with the huge amount of information, the real space is also shifting from the usual cloud system to the edge computing system for the Internet of Things (IoT). Then, what is machine learning suitable for edge computing? In this study, purpose of our research is to propose a system that executes learning while dividing data obtained from “things” into multiple servers. In other words, we propose and implement a system based on the idea of “division (distribution) of learning data + parallel computation = fast and secure learning method (suitable for edge computing)”.

Much research has been done on machine learning that supports artificial intelligence (AI). In particular, neural networks (NN) such as CNN have received much attention from theory to application. NN can easily answer “what to do” to many problems, but it is difficult to answer “why do that”. This is the so-called “black box problem” of AI. The fuzzy system is one of the systems that realizes machine learning and answers such problems. We are studying new learning methods for this system. For example, consider the following problem in which a moving object arrives at a destination avoiding obstacles. The usual machine learning method gives a route for the answer, but it is difficult to answer “why do that”. The fuzzy system gives the following qualitative rules as a result of learning.