Estimation of KL Divergence: Optimal Minimax Rate
Abstract
The problem of estimating the Kullback-Leibler divergence between two unknown distributions and is studied, under the assumption that the alphabet size of the distributions can scale to infinity. The estimation is based on independent samples drawn from and independent samples drawn from . It is first shown that there does not exist any consistent estimator that guarantees asymptotically small worst-case quadratic risk over the set of all pairs of distributions. A restricted set that contains pairs of distributions, with density ratio bounded by a function is further considered. {An augmented plug-in estimator is proposed, and its worst-case quadratic risk is shown to be within a constant factor of , if and exceed a constant factor of and , respectively.} Moreover, the minimax quadratic risk is characterized to be within a constant factor of , if and exceed a constant factor of and , respectively. The lower bound on the minimax quadratic risk is characterized by employing a generalized Le Cam's method. A minimax optimal estimator is then constructed by employing both the polynomial approximation and the plug-in approaches.
Cite
@article{arxiv.1607.02653,
title = {Estimation of KL Divergence: Optimal Minimax Rate},
author = {Yuheng Bu and Shaofeng Zou and Yingbin Liang and Venugopal V. Veeravalli},
journal= {arXiv preprint arXiv:1607.02653},
year = {2018}
}
Comments
IEEE Transactions on Information Theory