66
36

Subspace Embedding and Linear Regression with Orlicz Norm

Abstract

We consider a generalization of the classic linear regression problem to the case when the loss is an Orlicz norm. An Orlicz norm is parameterized by a non-negative convex function G:R+R+G:\mathbb{R}_+\rightarrow\mathbb{R}_+ with G(0)=0G(0)=0: the Orlicz norm of a vector xRnx\in\mathbb{R}^n is defined as xG=inf{α>0i=1nG(xi/α)1}. \|x\|_G=\inf\left\{\alpha>0\large\mid\sum_{i=1}^n G(|x_i|/\alpha)\leq 1\right\}. We consider the cases where the function G()G(\cdot) grows subquadratically. Our main result is based on a new oblivious embedding which embeds the column space of a given matrix ARn×dA\in\mathbb{R}^{n\times d} with Orlicz norm into a lower dimensional space with 2\ell_2 norm. Specifically, we show how to efficiently find an embedding matrix SRm×n,m<nS\in\mathbb{R}^{m\times n},m<n such that xRd,Ω(1/(dlogn))AxGSAx2O(d2logn)AxG.\forall x\in\mathbb{R}^{d},\Omega(1/(d\log n)) \cdot \|Ax\|_G\leq \|SAx\|_2\leq O(d^2\log n) \cdot \|Ax\|_G. By applying this subspace embedding technique, we show an approximation algorithm for the regression problem minxRdAxbG\min_{x\in\mathbb{R}^d} \|Ax-b\|_G, up to a O(dlog2n)O(d\log^2 n) factor. As a further application of our techniques, we show how to also use them to improve on the algorithm for the p\ell_p low rank matrix approximation problem for 1p<21\leq p<2.

View on arXiv
Comments on this paper