CP$^2$: Leveraging Geometry for Conformal Prediction via Canonicalization
This addresses the issue of unreliable uncertainty quantification in conformal prediction for applications like computer vision or robotics where geometric transformations are common, representing an incremental improvement by combining existing techniques.
The paper tackles the problem of conformal prediction failing under geometric data shifts like rotations or flips, and finds that integrating geometric information via pose canonicalization reinstates coverage guarantees and robustness, with evaluations showing broad applicability to black-box predictors.
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.