Return To The Sdm Model And Its Applied Research, Function Approximation And Identification

The Kanerva's Sparse distributed Memory (SDM) tackles the problem of training large data patterns and extendes the storage mode of existing computer .But it's address array produced randomly can't reveal the distribution of patterns and it has't the ability of function approximation for its learning rule. In this paper a Regressive SDM model (RSDM) is designed to resolve these problems.In RSDM, Binary patterns are replaced by real-valued patterns, accordingly avoiding the coding process; The outer learning rule is replaced by regression rule, therefore the model has not only the ability of pattern recognition but the ability of function approximation. The prearrangement of the address array bases on the distribution of patterns.If the distribution of patterns is uniform.then the address array is prearranged randomly, otherwise predisposed with the theory of genetic algorithm and the pruneing measure so as to indicate the distribution of patterns and improve the network performance.non-linear function approximation, time-series prediction and handwritten numeral recognition show that the modified model is effective and feasible.