| Abstract: |
| We will overview some matrix factorizations that allow one to observe only small randomly chosen submatrices of a data matrix, and how these factorizations can be applied in the design of fast algorithms for certain tasks such as Robust PCA or matrix completion. We show how to obtain state-of-the-art runtime for these tasks and apply the algorithms to some image and video processing tasks. We will discuss some natural generalizations of this approach to tensor data. |
|