Fast CVT-based Data Visualization Algorithms

GRAND Seminar 12:00 noon, April 13, Tue., 2010, ENGR 4201

Speaker

Maria Emelianenko
Assistant Professor
Department of Mathematical Sciences
George Mason University

Abstract

Efficient data visualization techniques are critical for many scientific applications. Centroidal Voronoi tessellation (CVT) based algorithms offer a convenient vehicle for performing image analysis, segmentation and compression while allowing to optimize retained image quality with respect to a given metric. In experimental science with data counts following Poisson distributions, several CVT-based data tessellation algorithms have been recently developed. Although they surpass their predecessors in robustness and quality of reconstructed data, time consumption remains to be an issue due to heavy utilization of the slowly converging Lloyd iteration. In this talk I will survey several possible approaches to accelerating CVT construction, including a recently developed multilevel optimization based CVT-based binning scheme. Its performance on a set of spectroscopy data, some convergence estimates and possible generalizations will be discussed.

Short Bio

Maria Emelianenko was born in Dubna, Russia. She received an M.S. in Applied Mathematics from Moscow State University and a Ph.D. in Mathematics from Pennsylvania State University in 2005. She spent two years as a Research Associate at Carnegie Mellon's Center for Nonlinear Analysis before joining George Mason University faculty in 2007. She is currently working on problems arising on the interface between mathematics, physics, biology and engineering.