2:00 PM - 4:00 PM
Room: Steamboat Hotel
High-dimensional data sets often exhibit low-dimensional structures. Traditional methods for recovering these structures fail for many important applications in imaging. The workshop will present emerging approaches for geometric data modeling, while targeting important imaging applications. The topics to be discussed include two alternatives to PCA (one robustly recovers low-rank approximation and the other accounts for higher-order cumulants), fast modeling of data with several affine subspaces (via a multiscale geometric procedure), and modeling data by mixtures of manifolds based on semi-supervised information. The applications include optical flow estimation from video, face recognition, motion segmentation, and identification of hand-written digits.
Organizer:
Gilad Lerman
University of Minnesota