NeurIPS 2020

How many samples is a good initial point worth in Low-rank Matrix Recovery?

Meta Review

This paper tries to quantify the effectiveness of good initial solution in the context of matrix factorization/sensing problems. More concretely, they showed that with a good initialization the RIP condition required for matrix factorization/sensing can be relaxed, resulting in requiring a smaller number of samples. Overall the reviewers found the perspective interesting and the results are clean.