分解データを用いる自律分散システムによる秘匿計算

抄録

With the progress of information technologies such as AI (Artificial Intelligence) and IoT (Internet of Things), research on machine learning using big data has been conducted. In some cases, complex and large-scale computations are required to realize machine learning, and in such cases, cloud and edge systems are widely used. On the other hand, in these systems, there is a risk of data loss or leakage by depositing data in an external server. For this reason, research on learning methods that maintain data security and are easy to use is widely conducted. As for secure computation methods using distributed processing, learning methods using subsets of data or decomposed data such as FL are well known. These methods are characterized by the fact that they achieve high confidentiality by repeating distributed processing, in which data and parameters in machine learning are distributed or decomposed and computed, and integrated at a central server. On the other hand, the role of the central server is important for the realization of machine learning by data distribution and integration. Therefore, it is desirable to realize machine learning by a decentralized autonomous method instead of such a centralized method.
In this paper, we propose a BP learning method for decentralized distributed systems using decomposed data and demonstrate its effectiveness.