Discriminative Subnetworks with Regularized Spectral Learning for Global-State Network Data

摘要: Data mining practitioners are facing challenges from data with network structure. In this paper, we address a specific class of global-state networks which comprises set instances sharing similar structure yet having different values at local nodes. Each instance is associated global state indicates the occurrence an event. The objective to uncover small discriminative subnetworks that can optimally classify values. Unlike most existing studies explore exponential subnetwork space, difficult problem by adopting space transformation approach. Specifically, present algorithm optimizes constrained dual-objective function learn low-dimensional subspace capable discriminating labelled states, while reconciling common topology across instances. Our takes appealing approach spectral graph learning and show globally optimum solution be achieved via matrix eigen-decomposition.