Modular neural networks for estimating odor concentrations

摘要: The concentration estimation for multiple kinds of odors is regarded first as two-class classification and then approximation problems, solved by single-output multi-layer perceptrons (MLPs) lined up in two parallel rows. A pair MLPs cascade on behalf a specified odor. n pairs represent odors, one one. An MLP the row separates its represented odor from others. Because training subsets are often unbalanced, samples minority sides virtually reinforced. generalization an limited local regions with respect to distribution second approximates relationship between responses sensor array concentrations sample assigned kind maximum output row, estimated another corresponding pair. effectiveness proposed models verified experiments 4 fragrant materials well their extended dataset.