1: \begin{proof}

2: Define the random variables $\{\mb{Y}_{M,i}; i = 1,\ldots,N\}$ for any fixed $M$ 

3: \begin{equation}

4: \mb{Y}_{M,i} = \frac{g({\mb{\tilde{f}}_k(\mb{X}_i)},\mb{X}_i) - \expect[g({\mb{\tilde{f}}_k(\mb{X}_i)},\mb{X}_i)]}{\sqrt{\var[g({\mb{\tilde{f}}_k(\mb{X}_i)},\mb{X}_i)]}}, \nonumber

5: \end{equation}

6: The key idea here is to recognize that $\mb{Y}_{M,i}$ are exchangeable random variables. Blum et.al. \cite{chernoff} showed that for exchangeable $0$ mean, unit variance random variables $\mb{Z_i}$, the sum $\mb{S}_N = \frac{1}{\sqrt{N}}\sum_{i=1}^N \mb{Z}_i$ converges in distribution to $N(0,1)$ if and only if $Cov(\mb{Z}_1,\mb{Z}_2) = 0$ and $Cov(\mb{Z}_1^2,\mb{Z}_2^2) = 0$. In our case, 

7: \begin{eqnarray}

8: Cov(\mb{Y}_{M,i},\mb{Y}_{M,j}) &=& O(1/M), \nonumber \\

9: Cov(\mb{Y}_{M,i}^2,\mb{Y}_{M,j}^2) &=& O(1/M). \nonumber 

10: \end{eqnarray}

11: As $M$ gets large, we then have that $Cov(\mb{Y}_{M,i},\mb{Y}_{M,j}) \to 0$ and $Cov(\mb{Y}_{M,i}^2,\mb{Y}_{M,j}^2) \to 0$. We then extend the work by Blum et.al. to show that convergence in distribution to $N(0,1)$ holds in our case as both $N$ and $M$ get large. These ideas are rigorously treated in Appendix E.

12: \end{proof}

13: