Tensor Regression and Neuroimaging Analysis
University of Michigan School of Public Health
3755 SPH I, 1415 Washington Heights Ann Arbor, MI 48109-2029
Classical regression models treat variables (predictor or response) as a vector and estimate a corresponding vector of regression coefficients. Modern applications in medical imaging generate variables of more complex form such as multidimensional arrays (tensors). Traditional statistical and computational methods are proving insufficient for analysis of those data due to their ultrahigh dimensionality as well as complex structure. In this talk, we propose a family of tensor predictor and response regression models that efficiently exploit the special structure of tensors. Under this framework, ultrahigh dimensionality is reduced to a manageable level, resulting in efficient estimation and prediction. Fast and highly scalable estimation algorithms are proposed, numerous forms of regularizations are studied, and asymptotic properties are obtained. Effectiveness of the new methods is demonstrated on real neuroimaging data analysis. Light refreshments for seminar guests will be served at 3:10 p.m. in 3755. Department of Biostatistics

Tensor Regression and Neuroimaging Analysis

Lexin Li, Ph.D., Associate Professor of Biostatistics - University of California-Berkeley

icon to add this event to your google calendarNovember 1, 2018
3:30 pm - 5:00 pm
3755 SPH I
1415 Washington Heights
Ann Arbor, MI 48109-2029
Sponsored by: Department of Biostatistics
Contact Information: Zhenke Wu (zhenkewu@umich.edu and Peisong Han (peisong@umich.edu)

Classical regression models treat variables (predictor or response) as a vector and estimate a corresponding vector of regression coefficients. Modern applications in medical imaging generate variables of more complex form such as multidimensional arrays (tensors). Traditional statistical and computational methods are proving insufficient for analysis of those data due to their ultrahigh dimensionality as well as complex structure. In this talk, we propose a family of tensor predictor and response regression models that efficiently exploit the special structure of tensors. Under this framework, ultrahigh dimensionality is reduced to a manageable level, resulting in efficient estimation and prediction. Fast and highly scalable estimation algorithms are proposed, numerous forms of regularizations are studied, and asymptotic properties are obtained. Effectiveness of the new methods is demonstrated on real neuroimaging data analysis. Light refreshments for seminar guests will be served at 3:10 p.m. in 3755.