Distances between Gaussian distributions

1 Wasserstein

A useful analytic result about Wasserstein-2 distance, i.e. \(W_2(\mu;\nu):=\inf\mathbb{E}(\Vert X-Y\Vert_2^2)^{1/2}\) for \(X\sim\nu\), \(Y\sim\mu\). Two Gaussians may be related thusly (Givens and Shortt 1984): \[\begin{aligned} d&:= W_2(\mathcal{N}(\mu_1,\Sigma_1);\mathcal{N}(\mu_2,\Sigma_2))\\ \Rightarrow d^2&= \Vert \mu_1-\mu_2\Vert_2^2 + \operatorname{tr}(\Sigma_1+\Sigma_2-2(\Sigma_1^{1/2}\Sigma_2\Sigma_1^{1/2})^{1/2}). \end{aligned}\]

In the centred case this the even simpler \[\begin{aligned} d&:= W_2(\mathcal{N}(0,\Sigma_1);\mathcal{N}(0,\Sigma_2))\\ \Rightarrow d^2&= \operatorname{tr}(\Sigma_1+\Sigma_2-2(\Sigma_1^{1/2}\Sigma_2\Sigma_1^{1/2})^{1/2}). \end{aligned}\]

2 Kullback-Leibler

Pulled from wikipedia:

\[ D_{\text{KL}}(\mathcal{N}(\mu_1,\Sigma_1)\parallel \mathcal{N}(\mu_2,\Sigma_2)) ={\frac {1}{2}}\left(\operatorname {tr} \left(\Sigma _{2}^{-1}\Sigma _{1}\right)+(\mu_{2}-\mu_{1})^{\mathsf {T}}\Sigma _{2}^{-1}(\mu_{2}-\mu_{1})-k+\ln \left({\frac {\det \Sigma _{2}}{\det \Sigma _{1}}}\right)\right).\]

In the centred case this reduces to

\[ D_{\text{KL}}(\mathcal{N}(0,\Sigma_1)\parallel \mathcal{N}(0, \Sigma_2)) ={\frac {1}{2}}\left(\operatorname{tr} \left(\Sigma _{2}^{-1}\Sigma _{1}\right)-k+\ln \left({\frac {\det \Sigma _{2}}{\det \Sigma _{1}}}\right)\right).\]

3 Hellinger

Djalil defines Hellinger distance \[\mathrm{H}(\mu,\nu) ={\Vert\sqrt{f}-\sqrt{g}\Vert}_{\mathrm{L}^2(\lambda)} =\Bigr(\int(\sqrt{f}-\sqrt{g})^2\mathrm{d}\lambda\Bigr)^{1/2}.\] via Hellinger affinity \[\mathrm{A}(\mu,\nu) =\int\sqrt{fg}\mathrm{d}\lambda, \quad \mathrm{H}(\mu,\nu)^2 =2-2A(\mu,\nu).\] For Gaussians it apparently turns out that \[\mathrm{A}(\mathcal{N}(m_1,\sigma_1^2),\mathcal{N}(m_2,\sigma_2^2)) =\sqrt{2\frac{\sigma_1\sigma_2}{\sigma_1^2+\sigma_2^2}} \exp\Bigr(-\frac{(m_1-m_2)^2}{4(\sigma_1^2+\sigma_2^2)}\Bigr),\]

In multiple dimensions: \[\mathrm{A}(\mathcal{N}(m_1,\Sigma_1),\mathcal{N}(m_2,\Sigma_2)) =\frac{\det(\Sigma_1\Sigma_2)^{1/4}}{\det(\frac{\Sigma_1+\Sigma_2}{2})^{1/2}} \exp\Bigr(-\frac{\langle\Delta m,(\Sigma_1+\Sigma_2)^{-1}\Delta m)\rangle}{4}\Bigr).\]

4 “Natural” / geodesic

High speed introduction to geometry of distances: Moakher and Batchelor (2006), which discusses the problems of comparing covariance matrices on a natural manifold.

The geodesic distance between \(\mathbf{P}\) and \(\mathbf{Q}\) in \(\mathcal{P}(n)\) is given by Lang (1999) \[ \mathrm{d}_R(\mathbf{P}, \mathbf{Q})=\left\|\log \left(\mathbf{P}^{-1} \mathbf{Q}\right)\right\|_F=\left[\sum_{i=1}^n \log ^2 \lambda_i\left(\mathbf{P}^{-1} \mathbf{Q}\right)\right]^{1 / 2}, \] where \(\lambda_i\left(\mathbf{P}^{-1} \mathbf{Q}\right), 1 \leq i \leq n\) are the eigenvalues of the matrix \(\mathbf{P}^{-1} \mathbf{Q}\). Because \(\mathbf{P}^{-1} \mathbf{Q}\) is similar to \(\mathbf{P}^{-1 / 2} \mathbf{Q} \mathbf{P}^{-1 / 2}\), the eigenvalues \(\lambda_i\left(\mathbf{P}^{-1} \mathbf{Q}\right)\) are all positive and hence [this] is well defined for all \(\mathbf{P}\) and \(\mathbf{Q}\) of \(\mathcal{P}(n)\).

5 Maximum mean discrepancy

With the Gaussian kernel, there is a closed-form expression for the distance of an RV from the standard Gaussian (Rustamov 2021).

The computation of the MMD requires specifying a positive-definite kernel; in this paper we always assume it to be the Gaussian RBF kernel of width \(\gamma\), namely, \(k(x, y)=e^{-\|x-y\|^2 /\left(2 \gamma^2\right)}\). Here, \(x, y \in \mathbb{R}^d\), where \(d\) is the dimension of the code/latent space, and we use \(\|\cdot\|\) to denote the \(\ell_2\) norm. The population MMD can be most straight-forwardly computed via the formula (Gretton et al. 2012): \[ \operatorname{MMD}^2(P, Q)=\mathbb{E}_{x, x^{\prime} \sim P}\left[k\left(x, x^{\prime}\right)\right]-2 \mathbb{E}_{x \sim P, y \sim Q}[k(x, y)]+\mathbb{E}_{y, y^{\prime} \sim Q}\left[k\left(y, y^{\prime}\right)\right] \]

[U]sing the sample \(Q_n=\left\{z_i\right\}_{i=1}^n\) of size \(n\), we replace the last two terms by the sample average and the U-statistic respectively to obtain the unbiased estimator: \[ \operatorname{MMD}_u^2\left(\mathcal{N}_d, Q_n\right)=\mathbb{E}_{x, x^{\prime} \sim \mathcal{N}_d}\left[k\left(x, x^{\prime}\right)\right]-\frac{2}{n} \sum_{i=1}^n \mathbb{E}_{x \sim \mathcal{N}_d}\left[k\left(x, z_i\right)\right]+\frac{1}{n(n-1)} \sum_{i=1}^n \sum_{j \neq i}^n k\left(z_i, z_j\right) \] Our main result is the following proposition whose proof can be found in Appendix C: Proposition. The expectations in the expression above can be computed analytically to yield the formula: \[ \operatorname{MMD}_u^2\left(\mathcal{N}_d, Q_n\right)=\left(\frac{\gamma^2}{2+\gamma^2}\right)^{d / 2}-\frac{2}{n}\left(\frac{\gamma^2}{1+\gamma^2}\right)^{d / 2} \sum_{i=1}^n e^{-\frac{\left\|z_i\right\|^2}{2\left(1+\gamma^2\right)}}+\frac{1}{n(n-1)} \sum_{i=1}^n \sum_{j \neq i}^n e^{-\frac{\left\|z_i-z_j\right\|^2}{2 \gamma^2}} \]

Extending this to arbitrary Gaussians, or analytic Gaussians, is left as an exercise.

Intuitively, we might prefer other kernels than the RBF if we are comparing Gaussians specifically; in particular the RBF is “wasteful” in that it controls moments of all orders, whereas we might only care about the first two moments (mean and covariance), for Gaussians.

6 References