Empirical Variance Minimization with Applications in Variance Reduction and Optimal Control

D. Belomestny,L. Iosipoi,Q. Paris,N. Zhivotovskiy
DOI: https://doi.org/10.48550/arXiv.1712.04667
2021-08-01
Abstract:We study the problem of empirical minimization for variance-type functionals over functional classes. Sharp non-asymptotic bounds for the excess variance are derived under mild conditions. In particular, it is shown that under some restrictions imposed on the functional class fast convergence rates can be achieved including the optimal non-parametric rates for expressive classes in the non-Donsker regime under some additional assumptions. Our main applications include variance reduction and optimal control.
Numerical Analysis,Machine Learning
What problem does this paper attempt to address?