non-negative matrix factorization is a basic tool for decomposing data into
the feature and weight matrices under non-negativity constraints, and in
practice is often solved in the alternating minimization framework. However, it
is unclear whether such algorithms can recover the ground