Applying SP-MLP to complex classification problems

摘要

A new model with special architecture, space-partitioning multi-layer perceptron (SP-MLP), is proposed in this paper to resolve some intractable classification problems. In the algorithm, the number of first hidden layer units of SP-MLP can be determined adaptively. Furthermore, we introduce a new sub-algorithm to improve the robustness of networks. The results of some simulation experiments show that SP-MLP is superior to other models. The issue of generalization is also discussed in the paper.