OnlineNMF.jl: A Julia Package for Out-of-core and Sparse Non-negative Matrix Factorization

Non-negative Matrix Factorization (NMF) is a widely used dimensionality reduction technique for identifying a small number of non-negative components that minimize the reconstruction error when applied to high-dimensional data (Meng, 2016;Stein-O'Brien, 2018).NMF has been applied across various fields of data science, including face recognition (Lee, 1999), audio signal processing (Kameoka, 2015), recommender system (Sajad, 2025), natural language processing (also known as a "topic model") (Srivastava & Sahami, 2009), population genetics (also known as "admixture analysis") (Simanovsky, 2019), and omics studies (Meng, 2016;Rodriques, 2019;Stein-O'Brien, 2018).Despite its broad applicability, NMF becomes computationally prohibitive for large data matrices, making it difficult to apply in practice.In particular, recent advances in single-cell omics have led to datasets with millions of cells, for which standard NMF implementations often fail to scale.To meet this requirement, I present OnlineNMF.jl,which is a Julia package to perform some NMF algorithms (https://github.com/rikenbit/OnlineNMF.jl).