A Self-adaptive Multi-hierarchical Modular Neural Network for Complex Problems

Due to the fact that the number of function models and the structure of the sub-model of the modular neural network are difficult to determine when applied to complex problems. This paper presents a self-adaptive multi-hierarchical modular neural network structure design method. In this method, a fast find of density peaks cluster algorithm is adopted to determine the number of the function modules, and a conditional fuzzy clustering algorithm is used to further divide the training samples of each function module into several groups to determine the number of sub-modules in each function module. For each sub-module, an incremental design of radical basis function (RBF) network network algorithm based on train error peak is applied to construct the structure of sub-modules which can self-adaptively build the structure of the sub-modules based on the training samples that allocated to the sub-modules. In sub-modules integration, a sub-module integrate approach based on relative distance measure is applied which can select different sub-modules from different function modules to collaboratively learning the training samples. Experiment results demonstrate that the self-adaptive multi-hierarchical modular neural network can not only solve the complex problems that the fully coupled RBF difficult to deal with, but also can improve the learning accuracy and generalization performance of the network.