23
60

Robust covariance estimation under L4L2L_4-L_2 norm equivalence

Abstract

Let XX be a centered random vector taking values in Rd\mathbb{R}^d and let Σ=E(XX)\Sigma= \mathbb{E}(X\otimes X) be its covariance matrix. We show that if XX satisfies an L4L2L_4-L_2 norm equivalence, there is a covariance estimator Σ^\hat{\Sigma} that exhibits the optimal performance one would expect had XX been a gaussian vector. The procedure also improves the current state-of-the-art regarding high probability bounds in the subgaussian case (sharp results were only known in expectation or with constant probability). In both scenarios the new bound does not depend explicitly on the dimension dd, but rather on the effective rank of the covariance matrix Σ\Sigma.

View on arXiv
Comments on this paper