Abstract:
|
We propose a sequential sparse representation classifier for robust vertex classification. This classifier, which is consistent for a variety of random graph models, represents a test vertex as a sparse combination of the training vertices and uses the regression coefficients to classify the test vertex. The proposed method first screens the training vertices using marginal regression, and then proceeds with $\ell 1$ minimization or orthogonal matching pursuit to derive the regression coefficients. This way, it achieves fast and robust performance without loss of accuracy compared to other benchmarks. We demonstrate the efficiency of the sequential classifier via simulations and real data experiments.
|