Local convergence of an algorithm for subspace identification with missing data

Start: 02/17/2015 - 10:00am
End  : 02/17/2015 - 11:00am

Applied Math Seminar

Laura Balzano (U of Michigan)


Low-dimensional linear subspace approximations to high-dimensional data find application in a great variety of applications where missing data are the norm, not only because of errors and failures in data collection, but because it may be impossible to collect and process all the desired measurements.

In this talk, I will describe recent results on estimating subspace projections from incomplete data. I will discuss the convergence guarantees and performance of the algorithm GROUSE (Grassmannian Rank-One Update Subspace Estimation), a subspace tracking algorithm that performs gradient descent on the Grassmannian. I will also discuss the relationship of GROUSE with an incremental SVD algorithm, and show results of GROUSE applied to problems in computer vision.

RN 104

Misc. Information