The Most Classic Algorithm for Non-Negative Matrix Factorization

This classic 1999 Nature publication introduced the most fundamental algorithm for non-negative matrix factorization (NMF). Despite its long-standing existence, this algorithm remains one of the most practical approaches in numerous applications. The implementation employs multiplicative update rules that iteratively optimize the objective function, ensuring non-negativity constraints are maintained throughout the factorization process. Its relatively straightforward implementation - typically involving matrix operations like element-wise multiplication and division - makes it an excellent starting point for learning matrix decomposition techniques. The core algorithm decomposes a given non-negative matrix V into two non-negative factors W and H, minimizing the reconstruction error through iterative updates. Over time, this foundational algorithm has been extensively refined and extended with variations including sparsity constraints, regularization terms, and alternative divergence measures to accommodate diverse application scenarios.