Prior specification is one of the hardest tasks in Bayesian modeling. In our new paper, we (Florence Bockting, @stefanradev.bsky.social and me) develop a method for expert prior elicitation using generative neural networks and simulation-based learning. arxiv.org/abs/2411.15826 - ThreadSky

paulbuerkner.com • 93 days ago

Prior specification is one of the hardest tasks in Bayesian modeling.

In our new paper, we (Florence Bockting, @stefanradev.bsky.social and me) develop a method for expert prior elicitation using generative neural networks and simulation-based learning.

https://arxiv.org/abs/2411.15826

Comments

sebastiendestercke.bsky.social•93 days ago

Was any real expert (meaning not a Bayesian) involved in the process, and if so how did you guarantee that the experts were not harmed in the process, as being normalized through flow can be pretty traumatic 😵‍💫?

The first part is a genuine question.

paulbuerkner.com•93 days ago

not at this stage. but we are planning follow up studies with real experts. the purpose of this paper was to develop the method. but we did try to keep the real experts already in mind.

kvwellstein.bsky.social•93 days ago

📌👀

nelsecostat.bsky.social•92 days ago

Do you have a more plain-language version of this abstract? I read it and skimmed the paper and I could not quickly figure out what problem you're attempting to solve or why such a sophisticated approach is needed.

paulbuerkner.com•92 days ago

I am sorry the abstract isnt so clear in that regard. we are currently updating our website of the related software and planning a tutorial paper, which should make this more clear.

nelsecostat.bsky.social•92 days ago

That's great. I work with a lot of practitioners and these kinds of approaches make them wary unless it's written in a way they can understand and explain to their colleagues, even when I'm the primary contact on the statistical aspects of the work.

francois-rozet.bsky.social•93 days ago

Interesting work! I am not familiar with the term prior explicitation, but after reading it seems related to the Empirical Bayes problem (which is to determine a prior over a latent variable given observations). How do they differ?

https://wikipedia.org/wiki/Empirical_Bayes_method

francois-rozet.bsky.social•93 days ago

Here are a few works that try to solve the Empirical Bayes problem using generative models.

https://arxiv.org/abs/1707.07113
https://arxiv.org/abs/2011.05836
https://arxiv.org/abs/2405.13712

paulbuerkner.com•92 days ago

Did you mean prior elicitation instead of explicitation?

paulbuerkner.com•92 days ago

prior eliciation is essentially transforming expert knowledge, provided in whatever form, into a mathematical presentation (our prior distribution) usable for inference. so it's not really empirical bayes since the latter tends to use the actual data not expert knowledge.

francois-rozet.bsky.social•92 days ago

Thank you, got it! Yes, I meant elicitation (I read that too fast, one could say I have a biased prior over words).

lacerbi.bsky.social•92 days ago

Nice work, another paper added to (the top of) the reading list! Sounds very relevant / related to our recent NeurIPS paper about prior elicitation with flows in the preferential setting. :) I shared this with my co-authors Petrus Mikkola & Arto Klami (neither of whom are around here).

paulbuerkner.com•92 days ago

thank you! we also cite your paper im our related work section. I previously discussed with Petrus and Arto about your and our approaches so they should be aware that the paper is coming eventually.

Comments

Posting Rules

Reply