Experimental parameter studies for the CMAC neural network

Summary form only given, as follows. Experimental and conceptual work done on the CMAC neural network is discussed. The behavior of the network in terms of training and generalization errors was investigated as a function of the various parameters available to the designer/user. Results include general principles of how to normalize the data for concise representation, evidence that the average capacity of a CMAC network is equal to the number of `available´ memory weights, an expression for the number of available weights, and plots of the normalized generalization error as a function of the number of training samples