SIC Seminar
Host: Moongu Jeon / Language: English
Thursday, December 24, 2015, 14:00-15:00,
#203, SIC-B Bldg. 2nd floor
Constrained Low Rank Approximations
for Scalable Data Analytics
Haesun Park, PhD
School of Computational Science and Engineering,
Georgia Institute of Technology, U.S.A.
Abstract:
Constrained low rank approximations have been widely utilized in large-scale data analytics where the applications reach far beyond the classical areas of scientific computing. We discuss some fundamental properties of nonnegative matrix factorization (NMF) and introduce some of its variants for clustering, topic discovery in text analysis, and community detection in social network analysis. In particular, we show how a simple rank 2 NMF combined with a divide-and-conquer framework results in a simple yet significantly more effective and scalable method for topic discovery. This simple approach can be further generalized for graph clustering and community detection. Substantial experimental results illustrate significant improvements both in computational time as well as quality of solutions obtained.
