David Sirl
[Joint Statistics & Probability and Mathematical Physics Seminar]
Low rank matrix recovery in high dimension : an iterative hard thresholding estimator with explicit limiting distribution
This talk will focus on a matrix recovery setting that is relevant for applications such as quantum tomography and matrix completion (e.g. the netflix challenge). The problem that will be considered is inference of the matrix given noisy observations. I will first provide an overview of the setting and its applications, as well as current literature. I will then discuss the problems of a high dimensional version of this setting, presenting challenges and existing results. Finally, I will introduce a new estimator based on iterative hard thresholding, that is minimax-optimal, computationally efficient, and has an explicit limiting distribution in a specific case. This presentation is based on joint work with Arlene K.Y. Kim, "An iterative hard thresholding estimator for low rank matrix recovery with explicit limiting distribution", arXiv:1502.04654.
The University of NottinghamUniversity Park Nottingham, NG7 2RD
For all enquiries please visit: www.nottingham.ac.uk/enquire