English

Estimation of KL Divergence: Optimal Minimax Rate

Information Theory 2018-02-22 v4 math.IT

Abstract

The problem of estimating the Kullback-Leibler divergence D(PQ)D(P\|Q) between two unknown distributions PP and QQ is studied, under the assumption that the alphabet size kk of the distributions can scale to infinity. The estimation is based on mm independent samples drawn from PP and nn independent samples drawn from QQ. 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 f(k)f(k) 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 (km+kf(k)n)2+log2f(k)m+f(k)n(\frac{k}{m}+\frac{kf(k)}{n})^2+\frac{\log ^2 f(k)}{m}+\frac{f(k)}{n}, if mm and nn exceed a constant factor of kk and kf(k)kf(k), respectively.} Moreover, the minimax quadratic risk is characterized to be within a constant factor of (kmlogk+kf(k)nlogk)2+log2f(k)m+f(k)n(\frac{k}{m\log k}+\frac{kf(k)}{n\log k})^2+\frac{\log ^2 f(k)}{m}+\frac{f(k)}{n}, if mm and nn exceed a constant factor of k/log(k)k/\log(k) and kf(k)/logkkf(k)/\log k, 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.

Keywords

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