抄録
Medical and health data associated with the use of AI to support a society with longevity
is highly sensitive, requiring learning methods through distributed data processing that achieve
privacy protection. In well-known secure distributed computation methods, such as Federated
Learning, a central server generally plays an important role in aggregating computations. However,
for more secure methods, it is desirable to realize machine learning using an autonomous decentralized
method that does not use a central server. This paper proposes a learning method with
confidentiality by autonomous distributed processing using decomposed data and parameters on
multiple server systems uniformly arranged in a ring structure. The advantage of the proposed
method is that the data and parameters can always be learned as decomposed data, thus protecting
security. In addition, the machine learning method can be implemented using a distributed processing
system that is easy to connect and has a uniform structure in which all servers perform the
same process, allowing for flexibility in responding to system changes and failures. Based on the
proposed method, we propose an algorithm for the Back Propagation method as an example of machine
learning application and show its effectiveness.
