A Sparse Infrastructure of Wavelet Network for Nonparametric Regression

Abstract

In this paper, we propose a novel 4-layer infrastructure of wavelet network. It differs from the commonly used 3-layer wavelet networks in adaptive selection of wavelet neurons based on the input information. As a result, it not only alleviates widespread structural redundancy, but can also control the scale of problem solution to a certain extent. Based on this architecture, we build a new type of wavelet network for function learning. The experimental results demonstrate that our model is remarkably superior to two well-established 3-layer wavelet networks in terms of both speed and accuracy. Another comparison to Huang’s real-time neural network shows that, at similar speed, our model achieves improvement in generalization performance. abstract environment.