Stay Informed:

COVID-19 (coronavirus) information
Zoom Links: Zoom Help | Teaching with Zoom | Zoom Quick Guide

Defense: Inference and Uncertainty Quantification for High-Dimensional Tensor Regression with Tensor Decompositions and Bayesian Methods

Speaker Name: 
Daniel Spencer
Speaker Title: 
Ph.D. Candidate
Start Time: 
Tuesday, May 26, 2020 - 9:00am
End Time: 
Tuesday, May 26, 2020 - 10:00am
Location: 
Zoom - https://ucsc.zoom.us/j/93404485322

Abstract: Certain image analysis settings rely on the use of large datasets with an inherent multidimensional data structures known as tensors. We explore different Bayesian modeling techniques for inference on tensor-valued coefficients with sparse, contiguous nonzero regions. We rely on the notion of tensor decomposition to reduce the parameter space within the Markov Chain Monte Carlo simulations, which improves inference while also preserving the spatial structures within the tensor coefficients. Bayesian shrinkage priors which match expected associations in the data are outlined and explained. The utility of these methods is demonstrated with applications in neuroimaging analysis.

Event Type: 
Adancement/Defense
Advisor: 
Rajarshi Guhaniyogi and Raquel Prado
Graduate Program: 
Statistical Science