Webinar Series: Mathematical Foundations of Data Science

Ultra-Sparse Models of Multiway Data

Speaker

Alfred O. Hero, University of Michigan  

Abstract

Modeling multi-way data is important for applications involving multi-indexed observables, e.g., hyperpsectral data that is indexed over spatial, frequency, and temporal dimensions.  The sparse matrix normal model is a multivariate Gaussian representation that expresses the covariance matrix as a Kronecker product of sparse lower dimensional covariances. This model is equivalent to assuming the conditional dependencies of the covariates  can be represented as a direct-product graph with few edges.   We will present an alternative framework based on Cartesian product graph representation and  Kronecker sums that leads to ultra-sparse models for multi-way data.   

Bio

Alfred Hero is the John H. Holland Distinguished University Professor of Electrical Engineering and Computer Science and the R. Jamison and Betty Williams Professor of Engineering at the University of Michigan.

Organizing Committee

Ethan X. Fang, Penn State University
Niao He, University of Illinois at Urbana-Champaign
Junwei Lu, Harvard University
Zhaoran Wang, NorthwesternUniversity
Zhuoran Yang, PrincetonUniversity
Tuo Zhao, Georgia Institute of Technology