Two dimensional function learning using CMAC neural network with optimized weight smoothing
Juan Vicente Pallotta, L.G. Kraft · 1999
This paper compares the traditional CMAC neural network weight update algorithm with a new optimized weight smoothing approach. Although CMAC learns functions rapidly, there is an inherent "roughness" to the approximation caused by spikes in the weight space even when the function being learned is relatively smooth. The new CMAC weight smoothing update scheme produces better approximations for a large class of functions.