SCIENCE@home

Online PCA with Optimal Regrets

Jiazhong Nie,W. Kotłowski,Manfred K. Warmuth

Published 2013 in International Conference on Algorithmic Learning Theory

ABSTRACT

We carefully investigate the online version of PCA, where in each trial a learning algorithm plays a k-dimensional subspace, and suffers the compression loss on the next instance when projected into the chosen subspace. In this setting, we give regret bounds for two popular online algorithms, Gradient Descent (GD) and Matrix Exponentiated Gradient (MEG). We show that both algorithms are essentially optimal in the worst-case when the regret is expressed as a function of the number of trials. This comes as a surprise, since MEG is commonly believed to perform sub-optimally when the instances are sparse. This different behavior of MEG for PCA is mainly related to the non-negativity of the loss in this case, which makes the PCA setting qualitatively different from other settings studied in the literature. Furthermore, we show that when considering regret bounds as a function of a loss budget, MEG remains optimal and strictly outperforms GD.

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-28 of 28 references · Page 1 of 1

CITED BY

Showing 1-35 of 35 citing papers · Page 1 of 1