Paper here, check it out! arxiv.org/abs/2411.01596

Special thanks to my coauthors, Claudio and Guilherme :)

In this paper we propose strategic conformal prediction, a simple extension of the standard conformal prediction methods which ensures that the uncertainty quantification is robust to strategic alterations.

Existing approaches to uncertainty quantification are great (especially conformal prediction!), but all break under strategic alterations -- i.e., when people try to alter the model's inputs in order to obtain particular undesirable outputs.

Typst is great, I use it for pretty much all my research notes. LaTeX only when preparing the submission.

Of course Bayesian inference can sometimes reproduce this for specific statistical models and priors, but that is extra work you have to do (showing that your Bayesian estimator has good frequentist properties; e.g., Bernsein-von-Mises)

Yes, but it's all a matter of what your inference means in the end. Bayesian inference only means that you have a 'valid belief' in general, whereas frequentist inference also means that you will be uncovering the 'true' parameter of the model.

Depends on the feed, no? 'Discover' has an algo, and even has 'show more like this' and 'show less like this' buttons in the dots menu

+1 for Quiet Posters. It's great.

Have you considered using Typst, though?

Design choices, usually yes. Language (i.e., macros) choices, absolutely not.

Heyo!

Oh hey, that's me on the right! Nice paper :)

Definitely into codegen, would love to chat!

🙋‍♂️