Mining dependence structures from statistical learning perspective

Abstract

Mining various dependence structures from data are important to many data mining applications. In this paper, several major dependence structure mining tasks are overviewed from statistical learning perspective, with a number of major results on unsupervised learning models that range from a single-object world to a multi-object world. Moreover, efforts towards a key challenge to learning have been discussed in three typical streams, based on generalization error bounds, Ockham principle, and BYY harmony learning, respectively.