A PROPOSAL OF PROFIT SHARING METHOD FOR SECURE MULTIPARTY COMPUTATION

Many studies for secure computation using shared data on the cloud system are made to avoid secure risks being abused or leaked and to reduce computing cost. The secure multiparty computation (SMC) is one of these methods. There are two methods for constructing a machine learning (ML) based on SMC. One is a method of sharing learning data into several subsets and learning at each server. The other method is to divide the learning data itself and learn by using each server. In the latter, we have proposed learning methods of BP, k-means and fuzzy inference about SMC so far. Further, we proposed a learning method of SMC on Q-learning which is one of reinforcement learning methods, and showed its effectiveness in the previous paper. Though Q-learning is a learning method with excellent generalization ability, it is known that it takes much time to learn. On the other hand, the profit sharing (PS) method is known to have a shorter learning time than Q-learning. Therefore, it is desired that PS learning method for SMC is superior in learning time to Q-learning method for SMC. In this paper, we propose PS learning method on SMC and show its effectiveness.