Optimal Subspace Estimation with Noise

Abstract

At this 5-minute lightning talk at the Discovery Institute, I will talk about my work on a project on approximating incomplete data with varieties. If there’s some unknown data that we think has linear structure and we only have access to noisy low dimensional projections (think - an unknown (linear) object in a dark room and I only show you the shadows after we shine a light), how accurately can we reconstruct the original unknown data? We derived an upper bound to this in our paper: a perturbation bound for the optimal subspace estimator from canonical projections

Karan Srivastava

PhD Student, Mathematics

My research interests include machine learning, reinforcement learning, combinatorics, and algebraic geometry