Robust Recovery of Low-rank Matrices and Tensors from Noisy Sketches


A common approach for compressing large-scale data is through matrix sketching. In this talk, we consider the problem of recovering low-rank matrices or tensors from noisy sketches. We provide theoretical guarantees characterizing the error between the output of the sketching algorithm and the ground truth low-rank matrix or tensor. Applications of this approach to synthetic data and medical imaging data will be presented.