English

Functional estimation in high-dimensional and infinite-dimensional models

Statistics Theory 2023-10-26 v1 Statistics Theory

Abstract

Let P{\mathcal P} be a family of probability measures on a measurable space (S,A).(S,{\mathcal A}). Given a Banach space E,E, a functional f:ERf:E\mapsto {\mathbb R} and a mapping θ:PE,\theta: {\mathcal P}\mapsto E, our goal is to estimate f(θ(P))f(\theta(P)) based on i.i.d. observations X1,,XnP,PP.X_1,\dots, X_n\sim P, P\in {\mathcal P}. In particular, if P={Pθ:θΘ}{\mathcal P}=\{P_{\theta}: \theta\in \Theta\} is an identifiable statistical model with parameter set ΘE,\Theta\subset E, one can consider the mapping θ(P)=θ\theta(P)=\theta for PP,P=Pθ,P\in {\mathcal P}, P=P_{\theta}, resulting in a problem of estimation of f(θ)f(\theta) based on i.i.d. observations X1,,XnPθ,θΘ.X_1,\dots, X_n\sim P_{\theta}, \theta\in \Theta. Given a smooth functional ff and estimators θ^n(X1,,Xn),n1\hat \theta_n(X_1,\dots, X_n), n\geq 1 of θ(P),\theta(P), we use these estimators, the sample split and the Taylor expansion of f(θ(P))f(\theta(P)) of a proper order to construct estimators Tf(X1,,Xn)T_f(X_1,\dots, X_n) of f(θ(P)).f(\theta(P)). For these estimators and for a functional ff of smoothness s1,s\geq 1, we prove upper bounds on the LpL_p-errors of estimator Tf(X1,,Xn)T_f(X_1,\dots, X_n) under certain moment assumptions on the base estimators θ^n.\hat \theta_n. We study the performance of estimators Tf(X1,,Xn)T_f(X_1,\dots, X_n) in several concrete problems, showing their minimax optimality and asymptotic efficiency. In particular, this includes functional estimation in high-dimensional models with many low dimensional components, functional estimation in high-dimensional exponential families and estimation of functionals of covariance operators in infinite-dimensional subgaussian models.

Keywords

Cite

@article{arxiv.2310.16129,
  title  = {Functional estimation in high-dimensional and infinite-dimensional models},
  author = {Vladimir Koltchinskii and Minghao Li},
  journal= {arXiv preprint arXiv:2310.16129},
  year   = {2023}
}
R2 v1 2026-06-28T13:00:44.162Z