Identify when combining WCP and NexCP weighting is useful

Determine problem settings in which combining the weighting mechanisms of weighted conformal prediction (likelihood-ratio weights addressing distribution shift) and nonexchangeable conformal prediction (index-swapping weights addressing nonexchangeability) would be useful for constructing conformal prediction sets.

Background

The paper presents a unified framework that encompasses several conformal prediction methods, including weighted conformal prediction (WCP) for known distribution shift and nonexchangeable conformal prediction (NexCP) for robustness under nonexchangeability. Although both are weighted approaches, they differ in their assumptions and technical motivations.

The authors note that their unified perspective suggests the potential to integrate these two weighting strategies. However, it remains unclear in which applications or data-generating scenarios such a combined approach would be advantageous. The explicit open question seeks to pinpoint those problem settings.

References

Our work provides a unified explanation for why they both work; this suggests the possibility of combining the two styles of weights, and raises an open question of determining problem settings in which such a combination might be useful.

— Unifying Different Theories of Conformal Prediction  (2504.02292 - Barber et al., 3 Apr 2025) in Discussion (Section 7)

Several directions remain open. The equivalent-radius and volume measures were used in this work to provide a practical normalization for comparing different Grassmannian metrics, but their extension to exact anisotropic prediction-set volumes in higher-dimensional Grassmannians would enable more precise efficiency comparisons. Locally anisotropic conformal regions whose geometry adapts not only in radius but also along different tangent directions, may further improve efficiency when uncertainty is strongly directional. Other extensions include alternative estimators of local geometric difficulty, data-driven selection or combination of Grassmannian metrics, and conformal methods designed for weaker forms of exchangeability or distribution shift.

— Adaptive conformal prediction on the Grassmannian  (2610.23508 - Loukrezis et al., 16 Oct 2026) in Section 7, Conclusions