Gradient Rectification for Robust Calibration under Distribution Shift

TL;DR

This paper introduces a novel calibration method for deep neural networks that improves reliability under distribution shifts by leveraging frequency-domain analysis and gradient-based rectification, without needing target domain data.

Contribution

It proposes a frequency-based filtering strategy combined with gradient rectification to enhance calibration robustness without access to target domain information.

Findings

Significantly improves calibration under distribution shift.

Maintains strong in-distribution calibration performance.

Effective on synthetic and real-world datasets like CIFAR-10/100-C and WILDS.

Abstract

Deep neural networks often produce overconfident predictions, undermining their reliability in safety-critical applications. This miscalibration is further exacerbated under distribution shift, where test data deviates from the training distribution due to environmental or acquisition changes. While existing approaches improve calibration through training-time regularization or post-hoc adjustment, their reliance on access to or simulation of target domains limits their practicality in real-world scenarios. In this paper, we propose a novel calibration framework that operates without access to target domain information. From a frequency-domain perspective, we identify that distribution shifts often distort high-frequency visual cues exploited by deep models, and introduce a low-frequency filtering strategy to encourage reliance on domain-invariant features. However, such information loss may degrade In-Distribution (ID) calibration performance. Therefore, we further propose a gradient-based rectification mechanism that enforces ID calibration as a hard constraint during optimization. Experiments on synthetic and real-world shifted datasets, including CIFAR-10/100-C and WILDS, demonstrate that our method significantly improves calibration under distribution shift while maintaining strong in-distribution performance.