Performance of Self-Generating Neural Network Applied to Pattern Recognition

摘要

Neural networks have been widely used in the field of intelligent information processing such as classification, clustering, learning, and recognition. However, the structures of neural networks are usually designed by human experts or their special experimental skill. Therefore, there exists a new problem how to construct a good structure to fit a given learning task. In order to avoid this situation, self-generating neural networks (SGNN) are recently proposed as an automatically constructing method from the given examples by Wen and Liu. The SGNN is implemented as self-generating neural tree (SGNT). In this paper, we present superior performances of SGNT when it is applied to character recognition problems. Basically, SGNT algorithm is generated as a kind of competitive learning algorithms. Consequently, it is natural to have a competent performance at the area of clustering or classification. However, our experimental results show that SGNT method is very efficient one to solve the pattern recognition problems, especially when they include a noisy signal problem.