Advances in Black-Box VI: Normalizing Flows, Importance Weighting, and Optimization

This popped up on Arxiv tonight. Looks like they benchmark a bunch of BBVI implementations against Stan’s ADVI, and of particular interest to the folks here, they use the Stan model repository for their benchmarks.


Oof. I know work is ongoing to set up posteriorDB as the gold-standard benchmarking model/result venue, but should we in the interim add a big disclaimer to the top of the old stan model repository that it should NOT be used for benchmarking? @avehtari @andrewgelman @breckbaldwin