Projection techniques to update the truncated SVD of evolving matrices with applications

Vasileios Kalantzis, Georgios Kollias, Shashanka Ubaru, Athanasios N. Nikolakopoulos, Lior Horesh, Kenneth Clarkson
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:5236-5246, 2021.

Abstract

This submission considers the problem of updating the rank-$k$ truncated Singular Value Decomposition (SVD) of matrices subject to the addition of new rows and/or columns over time. Such matrix problems represent an important computational kernel in applications such as Latent Semantic Indexing and Recommender Systems. Nonetheless, the proposed framework is purely algebraic and targets general updating problems. The algorithm presented in this paper undertakes a projection viewpoint and focuses on building a pair of subspaces which approximate the linear span of the sought singular vectors of the updated matrix. We discuss and analyze two different choices to form the projection subspaces. Results on matrices from real applications suggest that the proposed algorithm can lead to higher accuracy, especially for the singular triplets associated with the largest modulus singular values. Several practical details and key differences with other approaches are also discussed.

Cite this Paper


BibTeX
@InProceedings{pmlr-v139-kalantzis21a, title = {Projection techniques to update the truncated SVD of evolving matrices with applications}, author = {Kalantzis, Vasileios and Kollias, Georgios and Ubaru, Shashanka and Nikolakopoulos, Athanasios N. and Horesh, Lior and Clarkson, Kenneth}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {5236--5246}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/kalantzis21a/kalantzis21a.pdf}, url = {https://proceedings.mlr.press/v139/kalantzis21a.html}, abstract = {This submission considers the problem of updating the rank-$k$ truncated Singular Value Decomposition (SVD) of matrices subject to the addition of new rows and/or columns over time. Such matrix problems represent an important computational kernel in applications such as Latent Semantic Indexing and Recommender Systems. Nonetheless, the proposed framework is purely algebraic and targets general updating problems. The algorithm presented in this paper undertakes a projection viewpoint and focuses on building a pair of subspaces which approximate the linear span of the sought singular vectors of the updated matrix. We discuss and analyze two different choices to form the projection subspaces. Results on matrices from real applications suggest that the proposed algorithm can lead to higher accuracy, especially for the singular triplets associated with the largest modulus singular values. Several practical details and key differences with other approaches are also discussed.} }
Endnote
%0 Conference Paper %T Projection techniques to update the truncated SVD of evolving matrices with applications %A Vasileios Kalantzis %A Georgios Kollias %A Shashanka Ubaru %A Athanasios N. Nikolakopoulos %A Lior Horesh %A Kenneth Clarkson %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-kalantzis21a %I PMLR %P 5236--5246 %U https://proceedings.mlr.press/v139/kalantzis21a.html %V 139 %X This submission considers the problem of updating the rank-$k$ truncated Singular Value Decomposition (SVD) of matrices subject to the addition of new rows and/or columns over time. Such matrix problems represent an important computational kernel in applications such as Latent Semantic Indexing and Recommender Systems. Nonetheless, the proposed framework is purely algebraic and targets general updating problems. The algorithm presented in this paper undertakes a projection viewpoint and focuses on building a pair of subspaces which approximate the linear span of the sought singular vectors of the updated matrix. We discuss and analyze two different choices to form the projection subspaces. Results on matrices from real applications suggest that the proposed algorithm can lead to higher accuracy, especially for the singular triplets associated with the largest modulus singular values. Several practical details and key differences with other approaches are also discussed.
APA
Kalantzis, V., Kollias, G., Ubaru, S., Nikolakopoulos, A.N., Horesh, L. & Clarkson, K.. (2021). Projection techniques to update the truncated SVD of evolving matrices with applications. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:5236-5246 Available from https://proceedings.mlr.press/v139/kalantzis21a.html.

Related Material