An adaptive BYY harmony learning algorithm and its relation to rewarding and penalizing competitive learning mechanism

Under the Bayesian Ying-Yang (BYY) learning theory, a harmony function has been developed for a BI-architecture of the BYY system related to finite mixture model and its maximization leads to the parameter learning with automated model selection. This paper proposes an adaptive gradient algorithm for the harmony function. It has been further shown by theoretical analysis that the update rule of the algorithm is a kind of rewarding and penalizing competitive learning mechanism.