Predictive neural clustering system and its applications to bacteria properties predictions
Alexandr Alexandrovich Ezhov, V.A. Ilyin, L.A. Knizhnikova · 1992
A clustering criterion referred to as the Lakatos criterion is considered. A empty-class prediction approach is developed. Empty class representatives can be considered as predictions generated by the network. This interpretation is used in a predictive neural clustering system which can explore different core neural paradigms able to generate the empty classes. This system has been applied to microorganism clustering; specifically, it has been applied to the prediction of new forms of thermophilic bacteria.>