Dependence of CMAC Neural Network Properties at initial, during, and after Learning Phase from Input Mapping Function

CMAC neural network has only one adjustable weight layer. Its first layer is fixed and maps input space to association vector through receptive fields. Receptive field of standard CMAC is in shape of square function of given width. Lot of research now is done around different shapes for receptive fields, and combination with other connectionist structures. This paper will deal with the topics on, how will shape of receptive field influence approximation properties of CMAC, is there any relation between selected width and shape function for receptive field for better performance, and how will shape function of receptive field influence different phases, from initial, during, and after learning process. Key-Words: CMAC, receptive field shape functions, triangle receptive field, RBF receptive field, CMAC accuracy, CMAC generalization, receptive field width-shape function relation