Streaming Kernel PCA with $\tilde{O}(\sqrt{n})$ Random Features

Abstract

We study the statistical and computational aspects of kernel principal component analysis using random Fourier features and show that under mild assumptions, $O(\sqrt{n} \log n)$ features suffices to achieve $O(1/\epsilon^2)$ sample complexity. Furthermore, we give a memory efficient streaming algorithm based on classical Oja's algorithm that achieves this rate

Cite

Text

Ullah et al. "Streaming Kernel PCA with $\tilde{O}(\sqrt{n})$ Random Features." Neural Information Processing Systems, 2018.

Markdown

[Ullah et al. "Streaming Kernel PCA with $\tilde{O}(\sqrt{n})$ Random Features." Neural Information Processing Systems, 2018.](https://mlanthology.org/neurips/2018/ullah2018neurips-streaming/)

BibTeX

@inproceedings{ullah2018neurips-streaming,
  title     = {{Streaming Kernel PCA with $\tilde{O}(\sqrt{n})$ Random Features}},
  author    = {Ullah, Enayat and Mianjy, Poorya and Marinov, Teodor Vanislavov and Arora, Raman},
  booktitle = {Neural Information Processing Systems},
  year      = {2018},
  pages     = {7311-7321},
  url       = {https://mlanthology.org/neurips/2018/ullah2018neurips-streaming/}
}