Accelerating Nonnegative Matrix Factorization Over Polynomial Signals With Faster Projections

Nonnegative Matrix Factorization is a data analysis tool that aims at representing a set of input data vectors as nonnegative linear combinations of a few nonnegative basis vectors. When dealing with continuous input signals, smoothness and accuracy of this representation can often be improved if nonnegative polynomials are used in the basis instead of vectors. However, algorithms using polynomials are usually more computationally demanding than their vector counterparts.In this work, we consider the Hierarchical Alternating Least Squares method, which displays state-of-the art performance on this problem and requires at each iteration to compute projections over the set of nonnegative polynomials. We introduce several heuristic algorithms designed to provide fast approximations of these projections, and show that their use significantly accelerates the resolution of nonnegative matrix factorization problems over polynomial signals without adverse effect on the accuracy of the obtained solutions.

[1]  Ioannis Pitas,et al.  Nonnegative Matrix Factorization in Polynomial Feature Space , 2008, IEEE Transactions on Neural Networks.

[2]  Andrzej Cichocki,et al.  Nonnegative Matrix and Tensor Factorization T , 2007 .

[3]  Rafal Zdunek,et al.  Alternating direction method for approximating smooth feature vectors in Nonnegative Matrix Factorization , 2014, 2014 IEEE International Workshop on Machine Learning for Signal Processing (MLSP).

[4]  Andrzej Cichocki,et al.  Hierarchical ALS Algorithms for Nonnegative Matrix and 3D Tensor Factorization , 2007, ICA.

[5]  Patrik O. Hoyer,et al.  Non-negative Matrix Factorization with Sparseness Constraints , 2004, J. Mach. Learn. Res..

[6]  François Glineur,et al.  Nonnegative matrix factorization with polynomial signals via hierarchical alternating least squares , 2019, ESANN.

[7]  Lucas C. Parra,et al.  Nonnegative matrix factorization for rapid recovery of constituent spectra in magnetic resonance chemical shift imaging of the brain , 2004, IEEE Transactions on Medical Imaging.

[8]  B. Reznick,et al.  Polynomials that are positive on an interval , 2000 .

[9]  Victoria Stodden,et al.  When Does Non-Negative Matrix Factorization Give a Correct Decomposition into Parts? , 2003, NIPS.

[10]  Olga Taussky-Todd SOME CONCRETE ASPECTS OF HILBERT'S 17TH PROBLEM , 1996 .

[11]  W. M. Benzel,et al.  USGS Spectral Library Version 7 , 2017 .

[12]  Nurulfajar Abd Manap,et al.  β-Divergence Nonnegative Matrix Factorization on Biomedical Blind Source Separation , 2017 .

[13]  H. Sebastian Seung,et al.  Learning the parts of objects by non-negative matrix factorization , 1999, Nature.

[14]  Lieven De Lathauwer,et al.  Nonnegative Matrix Factorization Using Nonnegative Polynomial Approximations , 2017, IEEE Signal Processing Letters.

[15]  최승진 Algorithms for orthogonal nonnegative matrix factorization , 2018 .

[16]  Andrzej Cichocki,et al.  B-Spline Smoothing of Feature Vectors in Nonnegative Matrix Factorization , 2014, ICAISC.

[17]  D. Backenroth Methods in functional data analysis and functional genomics , 2018 .