Title:Generalized Categorization Axioms

Abstract:Categorization axioms have been proposed to bridge the difference between human recognition system and machine learning by an intuition: an object should be assigned to its most similar category. However, categorization axioms become trivial when the number of categories becomes one. In order to generalize categorization axioms into general cases, categorization input and categorization output are reinterpreted by inner and outer category representation. According to the categorization reinterpretation, two category representation axioms are presented. Category representation axioms and categorization axioms can combine into a generalized categorization axiomatic framework, which accurately delimit the theoretical categorization constraints. The proposed axiomatic framework reinterprets many results in machine learning in a unified way, such as regression, classification, clustering and dimensionality reduction.