Richard Zhao 501 Presentation

Title: Analysis of the Robustness of Nonnegative Matrix Factorization Algorithms

Abstract:
This literature review project presents the robustness of eliminating noises in facial recognition using the Nonnegative Matrix Factorization (NMF) algorithm with three different norms: L1-norm, L2-norm and L2,1-norm. We choose two benchmark image datasets, the AT\&T ORL dataset and the YaleB dataset, for numerical experiments with the above NMF algorithms. We examine the performance of these three different NMF methods to handle Gaussian noise based on three standard criteria: Relative Reconstruction Errors (RRE), Average Clustering Accuracy (ACC) and Normalized Mutual Information (NMI). It is observed that these three methods are comparable with each other for all the testing cases.