Nearly Optimal Clustering Risk Bounds for Kernel K-Means

Abstract

In this paper, we study the statistical properties of kernel $k$-means and obtain a nearly optimal excess clustering risk bound, substantially improving the state-of-art bounds in the existing clustering risk analyses. We further analyze the statistical effect of computational approximations of the Nyström kernel $k$-means, and prove that it achieves the same statistical accuracy as the exact kernel $k$-means considering only $\Omega (\sqrt{nk})$ Nyström landmark points. To the best of our knowledge, such sharp excess clustering risk bounds for kernel (or approximate kernel) $k$-means have never been proposed before.

References in corpus (7)

• 2004 IMS Medallion Lecture: Local Rademacher complexities and oracle inequalities in risk minimization
• Generalization Properties of Learning with Random Features
• Scalable Kernel K-Means Clustering with Nystrom Approximation: Relative-Error Bounds
• Prediction, Learning, Uniform Convergence, and Scale-sensitive Dimensions
• Learning Manifolds with K-Means and K-Flats
• Statistical and Computational Trade-Offs in Kernel K-Means
• ${\ell }_{\infty }$ Vector Contraction for Rademacher Complexity