CP$^2$: Leveraging Geometry for Conformal Prediction via Canonicalization

Putri A Van der Linden, Alexander Timans, Erik J Bekkers

Keywords: conformal prediction, geometry, canonicalization

TL;DR: We propose leveraging the canonicalization principle to supplement conformal prediction with geometric information, robustifying the procedure against geometric data shifts and ensuring fundamental conditions such as exchangeability are preserved.

Abstract: We study the problem of *conformal prediction* (CP) under geometric data shifts, where data samples are susceptible to transformations such as rotations or flips. While CP endows prediction models with *post-hoc* uncertainty quantification and formal coverage guarantees, their practicality breaks under distribution shifts that deteriorate model performance. To address this issue, we propose integrating geometric information—such as geometric pose—into the conformal procedure to reinstate its guarantees and ensure robustness under geometric shifts. In particular, we explore recent advancements on pose *canonicalization* as a suitable information extractor for this purpose. Evaluating the combined approach across discrete and continuous shifts and against equivariant and augmentation-based baselines, we find that integrating geometric information with CP yields a principled way to address geometric shifts while maintaining broad applicability to black-box predictors.

Code Link: https://github.com/computri/geometric_cp

Readers: auai.org/UAI/2025/Conference, auai.org/UAI/2025/Conference/Area_Chairs, auai.org/UAI/2025/Conference/Reviewers, auai.org/UAI/2025/Conference/Submission752/Authors, auai.org/UAI/2025/Conference/Submission752/Reproducibility_Reviewers

Submission Number: 752