Professor Steve Vavasis is the chair of the organizing committee of "Workshop on Optimization and Matrix Methods in Big Data" which will be held at the Fields Institute from February 9-13, 2015. The workshop is one of the events of the Thematic Program on Statistical Inference, Learning, and Models for Big Data.
Professor Vavasis works on convex optimization and its application to problems in data mining and machine learning. For example, he showed in 2009 that the widely used nonnegative matrix factorization is NP-hard. Despite this hardness result, he and Waterloo co-authors have found important special cases that can be solved efficiently with guaranteed accuracy using convex optimization. They have applied their methods to classifying features in image datasets and discovering material compositions in hyperspectral imaging. In collaboration with Waterloo graduate students Brendan Ames and Sahar Karimi, he has also developed new algorithms for large instances of convex optimization.
In the fall 2014 semester, Professor Vavasis will be teaching a graduate course on "optimization for big data".