Identifying Stable Components of Matrix /Tensor Factorizations via Low-Rank Approximation of Inter-Factorization Similarity

Many interesting matrix decompositions/factorizations, and especially many tensor decompositions, have to be solved by non-convex optimization-based algorithms, that may converge to local optima. Hence, when interpretability of the components is a requirement, practitioners have to compute the decomposition (e.g. CPD) many times, with different initializations, to verify whether the components are reproducible over repetitions of the optimization. However, it is non-trivial to assess such reliability or stability when multiple local optima are encountered. We propose an efficient algorithm that clusters the different repetitions of the decomposition according to the local optimum that they belong to, offering a diagnostic tool to practitioners. Our algorithm employs a graph-based representation of the decomposition, in which every repetition corresponds to a node, and similarities between components are encoded as edges. Clustering is then performed by exploiting a property known as cycle consistency, leading to a low-rank approximation of the graph. We demonstrate the applicability of our method on realistic electroencephalographic (EEG) data and synthetic data.

[1]  R. Bro PARAFAC. Tutorial and applications , 1997 .

[2]  Michal Mikl,et al.  Multiway Array Decomposition of EEG Spectrum: Implications of Its Stability for the Exploration of Large-Scale Brain Networks , 2017, Neural Computation.

[3]  Lieven De Lathauwer,et al.  Structured Data Fusion , 2015, IEEE Journal of Selected Topics in Signal Processing.

[4]  Rasmus Bro,et al.  Structure-revealing data fusion , 2014, BMC Bioinformatics.

[5]  Nico Vervliet,et al.  Tensorlab 3.0 — Numerical optimization strategies for large-scale constrained and coupled matrix/tensor factorization , 2016, 2016 50th Asilomar Conference on Signals, Systems and Computers.

[6]  Leonidas J. Guibas,et al.  Near-Optimal Joint Object Matching via Convex Relaxation , 2014, ICML.

[7]  Nikos D. Sidiropoulos,et al.  Tensor Decomposition for Signal Processing and Machine Learning , 2016, IEEE Transactions on Signal Processing.

[8]  Christian Jutten,et al.  Multimodal Data Fusion: An Overview of Methods, Challenges, and Prospects , 2015, Proceedings of the IEEE.

[9]  M. Hebert,et al.  Automatic three-dimensional modeling from reality , 2002 .

[10]  Leonidas J. Guibas,et al.  Consistent Shape Maps via Semidefinite Programming , 2013, SGP '13.

[11]  Lieven De Lathauwer,et al.  Coupled Canonical Polyadic Decompositions and (Coupled) Decompositions in Multilinear Rank- (Lr, n, Lr, n, 1) Terms - Part II: Algorithms , 2015, SIAM J. Matrix Anal. Appl..

[12]  Aapo Hyvärinen,et al.  Validating the independent components of neuroimaging time series via clustering and visualization , 2004, NeuroImage.

[13]  Sujay Sanghavi,et al.  Normalized Spectral Map Synchronization , 2016, NIPS.

[14]  Marc Pollefeys,et al.  Disambiguating visual relations using loop constraints , 2010, 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[15]  Qi-Xing Huang,et al.  SMAC: Simultaneous Mapping and Clustering Using Spectral Decompositions , 2018, ICML.