3
$\begingroup$

Suppose that we have a parametric distribution $P_{\theta}$, which is indexed by the parameter $\theta \in \mathbb{R}^d$. Let $W\{\cdot,\cdot\}$ be the Wasserstein Metric between two distributions.

What is the second-order expansion of $W\{P_{\theta+\Delta \theta}, P_{\theta}\}$ in terms of $\Delta \theta$ and the density of $P_{\theta}$? Moreover, what about the expansion of more general integral probability metrics (IPMs), for example, Dudley metric and Maximum Mean Discrepancy?

$\endgroup$
4
  • $\begingroup$ You have to be more specific about what kinds of IPM you have in mind. $\endgroup$
    – Henry.L
    Commented Aug 3, 2017 at 22:48
  • $\begingroup$ @Henry.L For example, Dudley metric and Maximum Mean Discrepancy. $\endgroup$
    – Minkov
    Commented Aug 4, 2017 at 0:48
  • $\begingroup$ Gigli has a Memoir of the AMS on second order calculus for W_2. It aims more at differential geometry, but might be relevant. $\endgroup$ Commented Aug 5, 2017 at 7:34
  • $\begingroup$ @BenoîtKloeckner Thanks a lot for the pointer! $\endgroup$
    – Minkov
    Commented Aug 7, 2017 at 7:03

0

You must log in to answer this question.