I’m hiring a postdoc to lead a major project at the intersection of *Diffusion Models*, *Bayesian Inference* and *Probabilistic Programming Languages*: Statistics Jobs - Statistics Forms. If interested, send me your CV and arrange for 2 letters of recommendation to be sent to me at aholbroo@g.ucla.edu.
Thanks for posting, @AndrewHolbrook. May I ask what the connection is to PPLs? I didn’t see more info in the job listing. I’m curious because I like programming languages and also because we have several postdocs who work on Bayesian inference with diffusions (Louis Grenioux, Mark Goldstein and Luhuan Wu [though she was technically visiting faculty on her way to Johns Hopkins] and have more on the way) as well as permanent research staff (Jiequn Han) getting into diffusions. We’ve also had one 30-person and one 100-person workshop on diffusion models. I’m excited about this because everyone tells me they’re better than RealNVP normalizing flows, and with the normalizing flows, Justin Domke was able to fit models that Stan can’t fit (like a hierarchal IRT-2PL model with lots of parameters and not much data), though it took a ridiculous number of flops to make it robust.
Hi @Bob_Carpenter, thanks for asking! This project has two main stages, each stage providing its own product. The first stage involves developing a semi-automated, agentic AI-based methodology for the large-scale expansion of posteriordb in a manner that guarantees model quality (only models from published works), sample quality, even distribution across application domains, and user accessibility through searchable tags/MD files. The second stage uses this database to build a foundation model that uses data and PPL scripts as context. Adding @mans_magnusson, with whom I’ve discussed this a little.