Open shoyer opened 7 years ago
yeah that would be great to have. we have operator vi in a separate code base (https://arxiv.org/abs/1610.09033), which is very related; we're hoping to merge operator vi in soon. in my experience i haven't found stein variational gradient descent to scale well to high dimensions (due to the kernel).
Is the codebase with operator VI publicly available? I've been looking, but couldn't find it.
https://arxiv.org/abs/1608.04471
It looks like an interesting inference technique that would be nice to have in Edward.