AI Paper Atlas

Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes

COMPASS: Robust Feature Conformal Prediction for Medical Segmentation Metrics

Matt Y. Cheung, Ashok Veeraraghavan, Guha Balakrishnan

ICLR 2026 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Conformal prediction methods

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
calibrate in feature space (COMPASS-J) instead of output-space SCPconformal prediction for segmentation area · Skin Lesion (alpha=0.10)interval size: 1813±127 -> 1179±53Improved
calibrate in feature space (COMPASS-J) instead of output-space SCPconformal prediction for segmentation area · PolyP (alpha=0.10)interval size: 6237±564 -> 4056±293Improved
use deeper features (COMPASS-J) instead of final logits (COMPASS-L) (ablation)COMPASS · PolyP (alpha=0.05)interval size: 6376±769 -> 5489±575Improved
use deeper features (COMPASS-J) instead of final logits (COMPASS-L) (ablation)COMPASS · H&E (alpha=0.05)interval size: 4408±432 -> 4637±630Hurt
use COMPASS instead of end-to-end CQR trained with Tversky losssegmentation interval · Skin Lesion (alpha=0.10)interval size: 1351±75 -> 1179±53Improved
add more principal components to COMPASS-J (ablation)COMPASS-J · segmentation areainterval tightnessNo clear effect
sum spatial dimensions of Jacobians before PCA (ablation)COMPASS-J · explained variance of Jacobiansexplained varianceImproved
use feature or Jacobian weights from deep layers in weighted CP (COMPASS-J)weighted COMPASS · H&E and Skin Lesion covariate shiftcoverageImproved
use class-label weights for weighted CPoutput-space CP · Skin Lesion covariate shiftcoverageHurt
use feature weighting with shallow-layer methods (COMPASS-L, E2E-CQR)weighted CP · H&E covariate shiftcoverageHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy