1: \begin{abstract}\noindent

2: We address the problem of estimating the \emph{difference} between

3: two probability densities.

4: A naive approach 

5: is a two-step procedure of first estimating two densities separately

6: and then computing their difference.

7: However, such a two-step procedure does not necessarily work well

8: because the first step is performed without regard to the second step

9: and thus a small error incurred in the first stage can cause a big error in the second stage.

10: In this paper, we propose a single-shot procedure 

11: for directly estimating the density difference

12: without separately estimating two densities.

13: We derive a non-parametric finite-sample error bound

14: for the proposed single-shot density-difference estimator

15: and show that it achieves the optimal convergence rate.

16: The usefulness of the proposed method is also demonstrated experimentally.

17: % We apply the proposed method to $L^2$-distance estimation and show that 

18: % it is more accurate than the difference of kernel density estimators

19: % and it is more robust against outliers than

20: % Kullback-Leibler divergence estimation.

21: 

22: \begin{center}

23:   \textbf{Keywords}

24: \end{center}

25: density difference,

26: $L^2$-distance,

27: robustness,

28: Kullback-Leibler divergence,

29: kernel density estimation.

30: %quadratic mutual information.

31: % Bregman divergence.

32: 

33: \end{abstract}

34: