AWM is excited to announce that we will be having our first seminar speaker next Monday, September 27th from 12-1 pm over Zoom! We are elated to have Dr. Misha Kilmer, Deputy Director of ICERM at Brown University and from Tufts University, come give us her talk on “The Case for Tensor Factorization: Compression, Analysis and Reconstruction of Image Data”. Hope to see you all there!
When: Monday September 27, 12-1pm
Where: Virtual on Zoom: https://wpi.zoom.us/j/96256127096
Speaker: Misha Kilmer, Tufts University, Deputy Director of ICERM at Brown University
Title: The Case for Tensor Factorization: Compression, Analysis, and Reconstruction of Image Data
Abstract: A grayscale digital image is a matrix (m x n array) whose entries represent intensity values. One way to compress an image, therefore, is to treat the image as a matrix for which a rank-revealing factorization, such as the singular value decomposition (SVD), can be computed. Vectors in the SVD corresponding to the largest singular values represent the directions with the most information content about the matrix. This information is retained and the rest of the vectors are discarded, providing an implicitly compressed representation of the image that retains the important features of the image for later analysis or reconstruction. However, a sequence or collection of k gray scale images (e.g. slices of a 3D medical image or frames in a video) can be arranged as an m x k x n data object. This multiway array is called a tensor and general concepts and algorithms from linear algebra are not easily generalizable to tensors. We introduce new methods for factoring such tensors which resemble matrix factorizations and illustrate how these can be used to successfully compress, analyze and reconstruct multiway image data. This talk is intended to accessible to students who have had a course in Linear Algebra.