CHEMNETS - THEORY AND APPLICATION

摘要

ChemNets are introduced for certain types of applications by taking advantage of previously developed chemical theories and incorporating them into neural network structures, Such a priori knowledge may help,in building a model that is one step closer to the true underlying model than a model constructed by a standard neural network. A robust and parsimonious neural structure with good predictive ability is expected for ChemNets. In this paper, the theory of ChemNets is presented, A ChemNet is designed for Taguchi sensors and compared to the standard neural network. The ChemNet constructed on the basis of Taguchi sensor theory showed a significant advantage in terms of parsimony, as noted in the relative parsimony of ChemNet model parameters in comparison with those of neural networks.