Recent SVD-free matrix factorization formulations have enabled rank minimization for systems with millions of rows and columns, paving the way for matrix completion for extremely high dimensional data. In this talk, we discuss robust matrix completion and subspace tracking algorithms that use factorized matrix decomposition with a pre-specified rank to detect and track a low rank subspace from incomplete measurements and in the presence of sparse noise. We demonstrate the performance of our algorithm for video background subtraction.
Hassan Mansour is a member of the research staff in the Multimedia Group at Mitsubishi Electric Research Laboratories, Cambridge, MA. He received his M.A.Sc. (2005) and Ph.D. (2009) from the Department of Electrical and Computer, University of British Columbia (UBC), Vancouver, Canada where he conducted research on scalable video coding and transmission. He then pursued a postdoctoral fellowship in the Departments of Mathematics, Computer Science, and Earth and Ocean Sciences at UBC working on theoretical and algorithmic aspects of compressed sensing and its application to seismic imaging.