Nonnegative Matrix Factorization and Its Applications to Pattern Analysis

Nonnegative Matrix Factorization and Its Applications to Pattern Analysis
Dr. Tolga Ensari
Department of Electrical and Computer Engineering
Computational Intelligence Lab
University of Louisville

 

4--5 PM, Davis Marksbury Theater, 329 Rose Street

Abstract: Finding a suitable representation of data is the main topic on data analysis research. There are several dimensionality reduction methods in the literature. Recently defined Nonnegative Matrix Factorization (NMF) is one of these methods, which has been applied for several tasks. NMF aims to find two nonnegative matrices (W and H) whose product can well approximate the given original data matrix (A). Its factorization can be shown with A≈W H equation. NMF, which allows reducing the dimension of data, is an unsupervised learning method. In this regard, the seminar firstly reviews how to obtain clustering of documents for documents (20-Newsgroups database) and face recognition data (ORL database). Additionally, it introduces the ‘correntropy’ entropy measure for minimizing the errors, which is also recently defined as a similarity measure between two random variables based on Information Theoretic Learning (ITL). The seminar also intends to explain that correntropy has better performance and extracts more information from the data structure than conventional approaches which use Euclidean distance. Finally, it will provide an overview of potential applications with NMF and correntropy-based NMF.


Brief Bio: Assistant Professor Ensari received B. Sc. and M. Sc. from Istanbul Technical University and PhD degree from Istanbul University, Turkey. His research area is neural networks and machine learning. Dr. Ensari was given an international successful researcher certificate for his research in 2006 and 2007 by Istanbul University. He worked in several research projects supported by TUBITAK and Istanbul University scientific research projects. Dr. Ensari has over 90 citations in the literature. He has been serving as a reviewer in several journals. He earned a research scholarship for his post doctoral study given by The Council of Higher Education (Turkish Government) in 2011. He studies at the University of Louisville as a research scholar and he is a member of IEEE.