Fast Nonnegative Matrix Factorization: An Active-Set-Like Method and Comparisons

By Jingu Kim, Haesun Park. SIAM Journal on Scientific Computing, 2011

353 citations recorded by the source corpus.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/92b18501fb7989ffa277805c42e0cd396dd2423b

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

On Tensors, Sparsity, and Nonnegative Factorizations (2011, neutral): “Much of the past work in nonnegative matrix and tensor analysis has focused on the LS error [44, 43, 6, 20, 26, 25, 23, 17, 27], which corresponds D ow nl oa de d 01 /0 2/ 13 to 1 98 .”
Fast and Efficient Estimation of Individual Ancestry Coefficients (2014, methodology): “For population genetic data, we found that alternating least-squares algorithms coupled with the active set method provided the best trade-oﬀ between speed and accuracy, and improved signiﬁcantly over other NMF implementations (Kim and Park 2011).”
UTOPIAN: User-Driven Topic Modeling Based on Interactive Nonnegative Matrix Factorization (2013, methodology): “For NMF, we have used one of the fastest and numerically reliable implementation based on an active set type of a least squares method [22].”
Generalized Low Rank Models (2014, methodology): “Hence a number of specialized algorithms and codes for fitting NNMF models are available [86, 91, 78, 80, 13, 79, 81].”
The Why and How of Nonnegative Matrix Factorization (2014, methodology): “…(ANLS) is a class of methods where the subproblems in W and H are solved exactly, that is, the update for W is given by Many methods can be used to solve the NNLS argmin W ≥ 0 || X − W H || F , and dedicated active-set methods have shown to perform very well in practice 6 ; see [72, 73, 75].”
SUSTain: Scalable Unsupervised Scoring for Tensors and its Application to Phenotyping (2018, methodology): “To assess the factors’ conciseness, we compare 3https://github.com/bdgp/staNMF their fit with the achieved fit of the real-valued model (NN CP-ALS), which is post-processed to achieve factor sparsity (as would be done by a practitioner).”
Outlier Detection for Text Data (2017, methodology): “H≥0f(W (k+1),H). (7) Since thesub-problemsare non-negativityconstrained least squares (NLS) problems, the two-block BCD method has been called thealternating non-negativeleast square (ANLS) framework [26, 19, 20]. 3.1.2 BCDwith 2k VectorBlocks- HALS/RRIMethod We partition the unknowns into 2k blocks”