Tag
The paper introduces Top-NMF, a framework that incorporates topological regularization via persistent homology into non-negative matrix factorization to learn interpretable and structurally coherent bases, applicable to spatially coherent image components, periodic time-series structures, and clique-like graph signals.