Neural networks with self-organized basis functions

In this study, we investigated a neural network structure that uses self-organized basis functions. The neural network is composed of submodules, each one consisting of several small associative memory blocks. The memory contents in these submodules are adjusted during the learning process in order to develop adequate basis functions. The neural network adds the outputs from these submodules to generate the network output. In a submodule, each associative memory block has a subset of the system inputs for forming the addresses. Each memory block is used to store some self-generated functions. The output of a submodule is the product of outputs from small memory blocks. The use of self-organized basis functions helps reduce the structure size, and the use of a subset of inputs to each memory block helps reduce the needed memory space