A method for reducing the amounts of training samples for developing AI systems

Abstract

A lot of tools are developed for AI (Artificial Intelligent) development. These tools are easy to use and the number of kinds of the tools are increasing quickly with new research results, therefore they are widely utilized for AI development in nowadays. A research issue here we need to solve is to provide methods for reducing training samples for AI development. The research issue comes from the background that most of the AI systems developed by using AI developing tools require a huge amount of training samples. In this paper, we propose a method for reducing the amount of training samples. Based on the proposed method, we created a Japanese hand-writing recognizing system to evaluate the effectiveness of the proposed method. This system is used for recognizing more than 6,000 different kinds of Japanese Kanji characters. The important feature of the system is that we do not need to collect millions of hand-writing Kanji character images as training samples. The effectiveness of the proposed method is confirmed by demonstration experiments.