Learning the parts of objects by non-negative matrix factorization¶

Why this mattered¶

TBD

(no abstract available)

cite → Maximum Likelihood from Incomplete Data Via the EM Algorithm — Lee and Seung use the EM algorithm as the optimization precedent for iterative maximum-likelihood-style updates in non-negative matrix factorization.
cite → Eigenfaces for Recognition — Lee and Seung contrast NMF's parts-based face representations with Eigenfaces' holistic PCA-based face representation.
cite ← Nonlinear Dimensionality Reduction by Locally Linear Embedding — Locally Linear Embedding relates to non-negative matrix factorization through the shared goal of unsupervised low-dimensional representation learning from high-dimensional data.
enables ← Maximum Likelihood from Incomplete Data Via the EM Algorithm — The EM algorithm supplied the iterative latent-variable estimation template used in probabilistic interpretations and multiplicative updates for non-negative matrix factorization.
enables ← Eigenfaces for Recognition — Eigenfaces showed that matrix factorization could learn compact face representations, while NMF replaced holistic PCA components with additive parts-based bases.