BYY binary independent systems, unsupervised harmony learning, and data mining of production rules

A concise systematic view on the Bayesian Ying-Yang (BYY) modular binary independent system is provided. The details on three typical architectures are discussed for both parameter learning and model selection, under the names of local binary factorial learning, local LMESR learning and competitive binary ICA learning, respectively. Then, they are applied to unsupervised knowledge mining via production rules with model selection that decides an appropriate number of rules, such that the redundancy and conflict among the rules can be reduced