1: \begin{abstract}

2: $f$-divergence estimation is an important problem in the fields of

3: information theory, machine learning, and statistics. While several

4: divergence estimators exist, relatively few of their convergence rates

5: are known. We derive the MSE convergence rate for a density plug-in

6: estimator of $f$-divergence. Then by applying the theory of optimally

7: weighted ensemble estimation, we derive a divergence estimator with

8: a convergence rate of $O\left(\frac{1}{T}\right)$ that is simple

9: to implement and performs well in high dimensions. We validate our

10: theoretical results with experiments.

11: \end{abstract}

12: