Uncertainty-Aware Post-Hoc Calibration: Mitigating Confidently Incorrect Predictions Beyond Calibration Metrics

TL;DR

The paper tackles the limitation of global calibration by proposing an uncertainty-aware post-hoc calibration framework that uses proximity-based conformal prediction to separate predictions into putatively correct and incorrect groups. It then applies a dual isotonic calibration: standard isotonic regression for the likely-correct group and underconfidence-regularized isotonic regression toward a uniform distribution for the likely-incorrect group, enabling explicit uncertainty signaling without changing predicted labels. Using MC-Dropout for posterior estimates and FAISS-based neighbor search, the method demonstrates reduced confidently incorrect predictions and competitive calibration metrics on CIFAR-10/100 with BiT and CoAtNet backbones, across multiple entropy thresholds. The approach emphasizes instance-level adaptivity, robustness to architecture, and a practical post-hoc workflow that improves both probability alignment and uncertainty-aware decision-making without model retraining, offering a meaningful step toward safer, more reliable AI systems.

Abstract

Despite extensive research on neural network calibration, existing methods typically apply global transformations that treat all predictions uniformly, overlooking the heterogeneous reliability of individual predictions. Furthermore, the relationship between improved calibration and effective uncertainty-aware decision-making remains largely unexplored. This paper presents a post-hoc calibration framework that leverages prediction reliability assessment to jointly enhance calibration quality and uncertainty-aware decision-making. The framework employs proximity-based conformal prediction to stratify calibration samples into putatively correct and putatively incorrect groups based on semantic similarity in feature space. A dual calibration strategy is then applied: standard isotonic regression calibrated confidence in putatively correct predictions, while underconfidence-regularized isotonic regression reduces confidence toward uniform distributions for putatively incorrect predictions, facilitating their identification for further investigations. A comprehensive evaluation is conducted using calibration metrics, uncertainty-aware performance measures, and empirical conformal coverage. Experiments on CIFAR-10 and CIFAR-100 with BiT and CoAtNet backbones show that the proposed method achieves lower confidently incorrect predictions, and competitive Expected Calibration Error compared with isotonic and focal-loss baselines. This work bridges calibration and uncertainty quantification through instance-level adaptivity, offering a practical post-hoc solution that requires no model retraining while improving both probability alignment and uncertainty-aware decision-making.

Figures (11)

• Figure 2: Reliability diagrams CIFAR-10 with BiT backbone.
• Figure 3: Reliability diagrams CIFAR-100 superclasses with BiT backbone.
• Figure 4: Reliability diagrams CIFAR-100 fine-grained classes with BiT backbone.
• Figure 5: Reliability diagrams CIFAR-10 with CoAtNet backbone.
• Figure 6: Reliability diagrams CIFAR-100 superclasses with CoAtNet backbone.