PROJECT TITLE :

A Unified Feature Selection Framework for Graph Embedding on High Dimensional Data

ABSTRACT:

Although graph embedding has been a powerful tool for modeling information intrinsic structures, merely employing all features for knowledge structure discovery might result in noise amplification. This can be significantly severe for top dimensional information with little samples. To satisfy this challenge, this paper proposes a completely unique efficient framework to perform feature choice for graph embedding, in that a category of graph embedding ways is cast as a least squares regression problem. In this framework, a binary feature selector is introduced to naturally handle the feature cardinality in the smallest amount squares formulation. The resultant integral programming problem is then relaxed into a convex Quadratically Constrained Quadratic Program (QCQP) learning downside, that will be efficiently solved via a sequence of accelerated proximal gradient (APG) ways. Since each APG optimization is w.r.t. solely a subset of options, the proposed method is quick and memory economical. The proposed framework is applied to several graph embedding learning problems, together with supervised, unsupervised, and semi-supervised graph embedding. Experimental results on many high dimensional information demonstrated that the proposed method outperformed the considered state-of-the-art ways.