|
April 17th, 12pm-1pm
Randomized Algorithms for Nonnegative Matrix FactorizationKoby Hayashi
Advisor: Rich Vuduc and Haesun Park
|
 |
ABSTRACT
Nonnegative Matrix Factorization (NMF) is a popular technique used in Machine Learning, Data mining, Computational Neuroscience, Image Segmentation, and more. We develop and analyze two randomized methods for computing NMF. Our methods are based on techniques from Randomized Numerical Linear Algebra. The first is based on computing a coarse grained factorization and then refining it. The second uses randomization to provably solve sequences of constrained least squares problems to a desired accuracy. Our methods result in up to 7.5x and 5.5x speedup on large dense and sparse problems. We validate the output of our randomized methods by testing them on two graph clustering tasks.
BIO
Koby Hayashi is a PhD Student in the College of Computing at Georgia Tech. He primarily works on low-rank approximation (LRA) methods for matrices and tensors. In this vein he has developed distributed memory and randomized algorithms for LRA methods. Additionally, he has explored LRA's applications to the problems of data clustering and latent feature discovery.