Nonnegative matrix factorization and I-divergence alternating minimization (Q2494663)

The non-negative matrix factorization problem can be described as follows: Given an elementwise non-negative matrix \(V \in \mathbb{R}_+^{m \times n}\) find, for assigned \(k\), non-negative matrices \(W \in \mathbb{R}_+^{m \times k}\) and \(H \in \mathbb{R}_+^{k \times n}\) such that \(V = WH\). Exact non-trivial non-negative factorizations do not always exist, see \textit{M. Hazewinkel} [CWI report PM-R8407 (May 1984)]. Hence for applications it is interesting to approximate factorization solutions. In the paper under review the authors study convergence properties of certain algorithms for finding such approximations.