作者: B.A. Whitehead , T.D. Choate
DOI: 10.1109/72.265957
关键词:
摘要: An evolutionary neural network training algorithm is proposed for radial basis function (RBF) networks. The locations of centers are not directly encoded in a genetic string, but governed by space-filling curves whose parameters evolve genetically. This encoding causes each group codetermined functions to fit region the input space. A produced from this evaluated its output connections only. Networks appear have better generalization performance on Mackey-Glass time series than corresponding networks determined k-means clustering. >