Prestructuring Neural Networks for Pattern Recognition Using Extended Dependency Analysis

摘要: We consider the problem of matching domain-specific statistical structure to neural-network (NN) architecture. In past work we have considered this in function approximation context; here pattern classification context. General Systems Methodology tools for finding problem-domain suffer exponential scaling computation with respect number variables considered. Therefore introduce use Extended Dependency Analysis (EDA), which scales only polynomially variables, desired analysis. Based on EDA, demonstrate a NN pre-structuring techniques applicable building neural classifiers. An example is provided EDA results significant dimension reduction input space, as well capability direct design an classifier.