100
155

Nonparametric empirical Bayes and compound decision approaches to estimation of a high-dimensional vector of normal means

Abstract

We consider the classical problem of estimating a vector \boldsμ=(μ1,...,μn)\bolds{\mu}=(\mu_1,...,\mu_n) based on independent observations YiN(μi,1)Y_i\sim N(\mu_i,1), i=1,...,ni=1,...,n. Suppose μi\mu_i, i=1,...,ni=1,...,n are independent realizations from a completely unknown GG. We suggest an easily computed estimator \boldsμ^\hat{\bolds{\mu}}, such that the ratio of its risk E(\boldsμ^\boldsμ)2E(\hat{\bolds{\mu}}-\bolds{\mu})^2 with that of the Bayes procedure approaches 1. A related compound decision result is also obtained. Our asymptotics is of a triangular array; that is, we allow the distribution GG to depend on nn. Thus, our theoretical asymptotic results are also meaningful in situations where the vector \boldsμ\bolds{\mu} is sparse and the proportion of zero coordinates approaches 1. We demonstrate the performance of our estimator in simulations, emphasizing sparse setups. In ``moderately-sparse'' situations, our procedure performs very well compared to known procedures tailored for sparse setups. It also adapts well to nonsparse situations.

View on arXiv
Comments on this paper