Affine Invariant Covariance Estimation for Heavy-Tailed Distributions
share
In this work we provide an estimator for the covariance matrix of a heavy-tailed random vector. We prove that the proposed estimator S admits affine-invariant bounds of the form (1-ε) SS (1+ε) Sin high probability, where S is the unknown covariance matrix, and is the positive semidefinite order on symmetric matrices. The result only requires the existence of fourth-order moments, and allows for ε = O(√(κ^4 d/n)) where κ^4 is some measure of kurtosis of the distribution, d is the dimensionality of the space, and n is the sample size. More generally, we can allow for regularization with level λ, then ε depends on the degrees of freedom number which is generally smaller than d. The computational cost of the proposed estimator is essentially O(d^2 n + d^3), comparable to the computational cost of the sample covariance matrix in the statistically interesting regime n ≫ d. Its applications to eigenvalue estimation with relative error and to ridge regression with heavy-tailed random design are discussed.
READ FULL TEXT
