My paper with David Blei and Edo Airoldi on Variational inference with copula augmentation is accepted into NIPS. The reviewer comments were very encouraging. Looks like higher quality posterior approximations are a worthy pursuit after all! We’ll add some clarifications and more experiment details in the camera-ready version.
As it is with all research, I’m already a little bored of that paper and very excited by new work which can be seen as its successor. I stand by the method as a great idea, and I hope we can solidify some of the new extensions soon. It will be made more scalable, more general, and also make headway for theoretical results on validating frequentist properties as a posterior density estimation technique(!).
Here’s the list of accepted NIPS papers: https://nips.cc/Conferences/2015/AcceptedPapers
From a quick keyword search on random things, there a lot of really cool papers I’m excited to read once the proceedings come out.