Learning structures of visual patterns from single instances

摘要

The main theme of this paper is learning. The basic idea of traditional similarity-based learning is that a program takes a number of instances, compares them in terms of similarities and differences, and describes the concept as a set of attributes common to positive instances. A concept seems to exist because it is necessary to discriminate the concept from other concepts. Therefore, it is just the attribute necessary to discriminate the concept from other concepts that must be learned. All attributes common to positive instances are not always necessary to describe the concept. That is, the existence of other concepts is very important to learn one concept. Inversely, the existence of the concept is important to other concepts too. Therefore, other concepts should be also learned simultaneously.In this paper, I propose a new learning method based on the differences among concepts. Character recognition is utilized as an example of learning.