SpatialUQ: Post-Hoc Uncertainty Quantification from Spatial Consistency in Black-Box Vision Models

Key points

• Clinical vision models are often deployed as frozen black boxes with no access to internals, retraining, or ground truth at inference time.1
• It measures the Jensen-Shannon divergence between the global prediction and the mean of five fixed spatial crops in six deterministic forward passes.1
• On NIH ChestX-ray14 (DenseNet-121, $N{=}25{,}596$), our Multicrop Uncertainty Score (MUS) reaches $0.784$ failure-detection AUC versus $0.664$ for MC-Dropout ($p{<}10^{-6}$) at one-fifth the compute, with native calibration ($SCE{=}0.049$ vs.\ $0.127$ for $\ell1$), the best-calibrated among methods above 0.78 AUC.1
• A supervised fusion of MUS with entropy, confidence, and $\ell1$ reaches $0.832$, outperforming a five-member ensemble ($0.813$).1

Sources (1)

• [1]SpatialUQ: Post-Hoc Uncertainty Quantification from Spatial Consistency in Black-Box Vision Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 05:53 AM

  We introduce SpatialUQ, a post-hoc uncertainty method using only output probabilities.

  Clinical vision models are often deployed as frozen black boxes with no access to internals, retraining, or ground truth at inference time.

Extractive summary: sentences quoted from the sources.