Ranked Entropy Minimization for Continual Test-Time Adaptation

TL;DR

This paper introduces ranked entropy minimization, a novel method for continual test-time adaptation that improves stability and prevents model collapse by structuring prediction difficulty and preserving entropy rank order.

Contribution

It proposes ranked entropy minimization to enhance stability in continual test-time adaptation, extending entropy minimization with a progressive masking strategy.

Findings

Effective across various benchmarks

Reduces model collapse in continual adaptation

Improves stability over traditional entropy minimization

Abstract

Test-time adaptation aims to adapt to realistic environments in an online manner by learning during test time. Entropy minimization has emerged as a principal strategy for test-time adaptation due to its efficiency and adaptability. Nevertheless, it remains underexplored in continual test-time adaptation, where stability is more important. We observe that the entropy minimization method often suffers from model collapse, where the model converges to predicting a single class for all images due to a trivial solution. We propose ranked entropy minimization to mitigate the stability problem of the entropy minimization method and extend its applicability to continuous scenarios. Our approach explicitly structures the prediction difficulty through a progressive masking strategy. Specifically, it gradually aligns the model's probability distributions across different levels of prediction difficulty while preserving the rank order of entropy. The proposed method is extensively evaluated across various benchmarks, demonstrating its effectiveness through empirical results. Our code is available at https://github.com/pilsHan/rem