Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging

2026-06-01

AI Summary

This work addresses the issue of model collapse in test-time adaptation, where entropy minimization—commonly used to encourage confident predictions—can induce prediction bias under distribution shifts, leading to feature cluster merging and catastrophic performance degradation. The study is the first to explicitly attribute this collapse to prediction bias and introduces DSBR, an unsupervised test-time bias rectification method that stabilizes adaptation by balancing each class’s contribution to the entropy loss. Notably, DSBR operates without requiring additional data or retraining. Extensive experiments on four medical imaging datasets and ImageNet-C demonstrate that DSBR effectively prevents model collapse and achieves performance on par with or superior to current state-of-the-art methods.

Abstract

Entropy minimization (EM) is the dominant objective for test-time adaptation, yet its failure mode, model collapse, remains poorly understood. In this work, we show that distribution shifts can cause feature clusters corresponding to distinct classes in the model's representation space to merge, while the decision boundary remains fixed. This induces a systematic skew in the predicted class distribution, referred to as prediction bias. Prediction bias refers to a shift in the predicted class distribution, with some classes overrepresented and others suppressed. We show that entropy minimization amplifies this prediction bias by tightening the existing clusters, reinforcing the incorrect groupings until all predictions collapse to a trivial solution. Next, to demonstrate the significance of prediction bias and mitigate it, we further propose Distribution Shift Bias Reduction (DSBR), a bias-correcting objective that specifically targets this failure mode by equalizing the contribution of each predicted class to the unsupervised entropy minimization loss. To study this failure mode, we design suitable adaptation settings using four medical-imaging datasets and additionally evaluate on ImageNet-C. We find that DSBR consistently stabilizes test-time adaptation, prevents model collapse, and matches or outperforms state-of-the-art methods. Moreover, DSBR operates solely at test-time.

Problem

• entropy minimization
• model collapse
• prediction bias
• distribution shift
• test-time adaptation

Innovation

• entropy minimization
• model collapse
• prediction bias
• test-time adaptation
• distribution shift