When and Where to Reset Matters for Long-Term Test-Time Adaptation
This addresses the issue of suboptimal adaptation and knowledge loss in continual test-time adaptation for machine learning practitioners, though it is incremental as it builds on existing reset strategies.
The paper tackles the problem of model collapse in long-term test-time adaptation, where errors accumulate and cause the model to predict only a few classes, by proposing an Adaptive and Selective Reset scheme that dynamically determines when and where to reset, along with a regularizer and adjustment scheme, achieving effectiveness in benchmarks under challenging conditions.
When continual test-time adaptation (TTA) persists over the long term, errors accumulate in the model and further cause it to predict only a few classes for all inputs, a phenomenon known as model collapse. Recent studies have explored reset strategies that completely erase these accumulated errors. However, their periodic resets lead to suboptimal adaptation, as they occur independently of the actual risk of collapse. Moreover, their full resets cause catastrophic loss of knowledge acquired over time, even though such knowledge could be beneficial in the future. To this end, we propose (1) an Adaptive and Selective Reset (ASR) scheme that dynamically determines when and where to reset, (2) an importance-aware regularizer to recover essential knowledge lost due to reset, and (3) an on-the-fly adaptation adjustment scheme to enhance adaptability under challenging domain shifts. Extensive experiments across long-term TTA benchmarks demonstrate the effectiveness of our approach, particularly under challenging conditions. Our code is available at https://github.com/YonseiML/asr.