Generalization and Informativeness of Conformal Prediction

Add to your list(s) Download to your calendar using vCal

• Dr Fredrik Hellstrom, UCL
• Wednesday 05 February 2025, 14:00-15:00
• MR5, CMS Pavilion A.

If you have a question about this talk, please contact Prof. Ramji Venkataramanan.

A popular technique for uncertainty quantification is conformal prediction, which converts point predictions into set predictions that are guaranteed to contain the true label of a test input with a user-defined probability. However, the size of the predicted set—-and thus the informativeness of the prediction—-is not controlled. In this talk, we present a theoretical connection between the informativeness of conformal prediction sets and generalization properties of the underlying model. Furthermore, we extend this analysis to conformal risk control and covariate shifts. The results provide insight into the effect of task-specific quantities and algorithmic hyperparameters, which we also illustrate via experiments.

This talk is part of the Information Theory Seminar series.

This talk is included in these lists:

• All CMS events
• All Talks (aka the CURE list)
• CMS Events
• DPMMS Lists
• DPMMS info aggregator
• DPMMS lists
• Hanchen DaDaDash
• Information Theory Seminar
• Interested Talks
• MR5, CMS Pavilion A
• School of Physical Sciences
• Statistical Laboratory info aggregator
• bld31

Note that ex-directory lists are not shown.