Stochastic Approximation for Canonical Correlation Analysis
Abstract
We propose novel first-order stochastic approximation algorithms for canonical correlation analysis (CCA). Algorithms presented are instances of inexact matrix stochastic gradient (MSG) and inexact matrix exponentiated gradient (MEG), and achieve $\epsilon$-suboptimality in the population objective in $\operatorname{poly}(\frac{1}{\epsilon})$ iterations. We also consider practical variants of the proposed algorithms and compare them with other methods for CCA both theoretically and empirically.
Cite
Text
Arora et al. "Stochastic Approximation for Canonical Correlation Analysis." Neural Information Processing Systems, 2017.Markdown
[Arora et al. "Stochastic Approximation for Canonical Correlation Analysis." Neural Information Processing Systems, 2017.](https://mlanthology.org/neurips/2017/arora2017neurips-stochastic/)BibTeX
@inproceedings{arora2017neurips-stochastic,
title = {{Stochastic Approximation for Canonical Correlation Analysis}},
author = {Arora, Raman and Marinov, Teodor Vanislavov and Mianjy, Poorya and Srebro, Nati},
booktitle = {Neural Information Processing Systems},
year = {2017},
pages = {4775-4784},
url = {https://mlanthology.org/neurips/2017/arora2017neurips-stochastic/}
}