Conformal Prediction as Bayesian Quadrature
Jake C. Snell, Thomas L. Griffiths · 2025
Read Conformal Prediction as Bayesian Quadrature by Jake C. Snell, Thomas L. Griffiths (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-13228.txt.
Description
Jake Snell and Thomas Griffiths rethink a popular safety method for AI forecasts in their 2025 paper. Conformal prediction gives a set of likely answers with a promise about error rates, even when the model is a black box. But its promises rest on frequentist probability, which can limit where the method makes sense.
The authors recast the method's core steps in Bayesian terms, which track belief rather than long-run frequency. Seen this way, some gaps in the usual promises stand out. They then offer a practical fix based on Bayesian quadrature, a way to estimate ranges by smart sampling. Their version gives clearer promises and a richer picture of the losses a user may face at test time.
Catalog record
| CREATOR | Jake C. Snell, Thomas L. Griffiths |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-13228