Beyond Prediction-Set Size: Quantifying Conformal Uncertainty through Credal Geometry

Name: Beyond Prediction-Set Size: Quantifying Conformal Uncertainty through Credal Geometry
Uploaded: 2026-10-15T22:31:22.364713+00:00
Duration: 346 s
Description: The work introduces Maximum Mean Imprecision (MMI-CP) to measure epistemic predictive uncertainty in conformal prediction by analyzing conflicting information within the induced credal set. This method provides a more informative assessment than just using prediction set size, showing MMI-based approaches outperform size-based methods in active learning and classification.

The work introduces Maximum Mean Imprecision (MMI-CP) to measure epistemic predictive uncertainty in conformal prediction by analyzing conflicting information within the induced credal set. This method provides a more informative assessment than just using prediction set size, showing MMI-based approaches outperform size-based methods in active learning and classification.

Open the full episode