Conformalized Uncertainty-Aware Framework for Trustworthy Cancer Subtyping in Whole-Slide Images
A model-agnostic wrapper that makes a digital-pathology classifier trustworthy by combining three components: a spectral-normalized neural Gaussian process that produces uncertainty-aware representations and flags out-of-scope inputs, an ambiguity-guided elimination of noisy image tiles, and conformal prediction that returns label sets with a statistically guaranteed error rate. Students learn how data-side and model-side reliability are separated and how each is enforced without retraining the underlying model.
2501.00053
Ensuring trustworthiness is fundamental to the development of artificial intelligence (AI) that is considered societally responsible, particularly in cancer diagnostics, where a misdiagnosis can have…