Reviews: Recovery Guarantee of Non-negative Matrix Factorization via Alternating Updates

The recovery guarantee for NMF with mild conditions is very important for the machine learning community. This paper provides a solution under three main conditions as follows. However, it is unclear if this initialization can be achieved. The proof technique seems to be novel, i.e., different from the popular techniques used for alternating minimization of matrix completion or the tensor methods. Therefore, it will probably be useful for the community.