Approaches to estimating local and global covariance and reparameterizing

There’s been a lot of off-Discourse discussion of this paper, which was first brought up a while ago in this topic:

There’s another paper on adapting the geometry that I’d like to add to the discussion:

There are bound to be more of these, which is why I wanted to open a thread to collect them. There are lots of references in the Leimkuhler et al. paper.


Dear Bob,

Here is another one, somewhat of the same flavour as DRHMC, but based on importance sampling/Laplace approximations, fresh on arXiv today. The main takeaway is that only crude “approximations to the Laplace approximation” seem to be sufficient/optimal.