Self-training with Consensus Making Mechanism for Artificial Neural Networks

— Analysis of Training Results with Naïve Implementation —

Abstract

Pattern recognition and event detection are essential technique for man-machine symbiosis. Supervised learning is one of the promising technique to achieve such ability, however gathering a large amount of training data is an extremely bothersome issue for its application. To tackle this problem, this study deals with self-training method which uses unknown input data for training with self-predicted output. This study particularly focuses on the self-prediction part in self-training, and proposes a novel self-prediction method based on collective modulation mechanism on consensus making. Through the numerical experiment with a naïve implementation, this report discusses the requirements and improvement points of the proposed mechanism.