Complex decision boundaries using tree connected single layer nets

The neural tree classifier architecture presented is a combination of two pattern recognition methods: tree classifiers and neural networks. Multilayer neural networks have the ability to create nonlinear classification boundaries when used for pattern recognition. These nets suffer from slow training times as well as high computational requirements. Single-layer nets do not have these disadvantages but cannot provide nonlinear decision boundaries. The authors present an approach that uses a series of single layer nets arranged in a binary-tree structure to approximate the performance of multilayer nets. By properly dividing the input space into tree-controlled clusters and assigning a single layer net to each cluster this approach can handle any classification task where multilayer nets have previously been used