Workshop notes: https://github.com/elcorto/37c3_uq_meetup
Uncertainty quantification (UQ) methods enable us to equip machine learning model predictions with "error bars". These are useful in many areas from active learning to out-of-distribution detection and in general to tackle trustworthiness issues (think driving, health, legal applications).
We'll also touch on recent methods to make generative models (hello ChatGPT) express confidence in their answers (i.e. is the anser likely to be true or a hallucination?).
I'll give an introduction to the main ideas and most common methods, followed by a relaxed open discussion.
This will be a casual meetup for people who apply and/or are interested in uncertainty quantification (UQ) for machine learning models. Whether you are an expert or have never heard of UQ, let's talk!
This won't be a talk or hands-on workshop, just an exchange of ideas. Which UQ methods have you used so far? You haven't, but like to learn where to start? Great, then join us!