English

Sparse Hanson-Wright inequalities for subgaussian quadratic forms

Probability 2017-02-21 v3 Statistics Theory Statistics Theory

Abstract

In this paper, we provide a proof for the Hanson-Wright inequalities for sparsified quadratic forms in subgaussian random variables. This provides useful concentration inequalities for sparse subgaussian random vectors in two ways. Let X=(X1,,Xm)RmX = (X_1, \ldots, X_m) \in \mathbb{R}^m be a random vector with independent subgaussian components, and ξ=(ξ1,,ξm){0,1}m\xi =(\xi_1, \ldots, \xi_m) \in \{0, 1\}^m be independent Bernoulli random variables. We prove the large deviation bound for a sparse quadratic form of (Xξ)TA(Xξ)(X \circ \xi)^T A (X \circ \xi), where ARm×mA \in \mathbb{R}^{m \times m} is an m×mm \times m matrix, and random vector XξX \circ \xi denotes the Hadamard product of an isotropic subgaussian random vector XRmX \in \mathbb{R}^m and a random vector ξ{0,1}m\xi \in \{0, 1\}^m such that (Xξ)i=Xiξi(X \circ \xi)_{i} = X_{i} \xi_i, where ξ1,,ξm\xi_1, \ldots,\xi_m are independent Bernoulli random variables. The second type of sparsity in a quadratic form comes from the setting where we randomly sample the elements of an anisotropic subgaussian vector Y=HXY = H X where HRm×mH \in \mathbb{R}^{m\times m} is an m×mm \times m symmetric matrix; we study the large deviation bound on the 2\ell_2-norm of DξYD_{\xi} Y from its expected value, where for a given vector xRmx \in \mathbb{R}^m, DxD_{x} denotes the diagonal matrix whose main diagonal entries are the entries of xx. This form arises naturally from the context of covariance estimation.

Keywords

Cite

@article{arxiv.1510.05517,
  title  = {Sparse Hanson-Wright inequalities for subgaussian quadratic forms},
  author = {Shuheng Zhou},
  journal= {arXiv preprint arXiv:1510.05517},
  year   = {2017}
}

Comments

29 pages; added full proof of Theorem 1.2 using moment generating functions, which had appeared in TR 539, October 2015

R2 v1 2026-06-22T11:23:42.519Z