Statistics and Biostatistics seminar series
Room: M3 3127
Tensor Regressions: New Methods, Algorithms and Theory
Classical regression models focus on variables, predictor or response, that are vectors and estimate a vector of regression coefficients. Modern applications in medical imaging and business generate variables of more complex forms such as multidimensional arrays (or tensors). Traditional statistical and computational methods are insufficient for the analysis of those data due to their high dimensionality as well as complex structures. In this talk, we discuss new tensor predictor and response regression models, highlight challenges in their estimation and theoretical analysis, and present some solutions. Efficacy of the newly proposed methods is demonstrated through the analyses of medical imaging and advertisement data.