发表机构
Poznan University of Technology; Graz University of Technology; IDEAS Research Institute; Computer Vision Center, Universitat Autonoma de Barcelona; NASK - National Research Institute; Gdańsk University of Technology（波兹南理工大学; 格拉茨理工大学; IDEAS研究院; 巴塞罗那自治大学计算机视觉中心; NASK国家研究院; 格但斯克理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究针对测试时适应常用的师生框架，发现指数移动平均教师权重策略仍存在误差累积，提出采用不更新权重的强硬教师，该改动可提升TTA方法在长场景数据集上的性能与超参数鲁棒性，且适用于语义分割等多种设置。
AI 中文摘要
测试时适应（Test-Time Adaptation, TTA）是近年来兴起的一种有前景的策略，可在部署时让预训练模型适应不断变化的数据分布，且无需任何标签。为缓解误差累积，研究者广泛采用师生框架，但该框架的长期稳定性常被视为理所当然。本研究通过实验表明，将教师权重设为学生权重的指数移动平均（exponential moving average）这一常用策略仍会出现误差累积，只是该现象在较长序列上比常用序列更明显。我们分析了师生框架内的稳定性-可塑性权衡，提出使用不更新权重的“强硬教师（intransigent teacher）”。令人惊讶的是，这一简单改动能让TTA方法在多个含长场景的数据集上显著提升性能，还能增强对超参数变化的鲁棒性。最后，我们证明这些改动可无缝且有效地应用于多种架构和实验设置，包括语义分割。代码可在该https URL获取。
英文摘要
Test-Time Adaptation (TTA) has recently emerged as a promising strategy that allows the adaptation of pre-trained models to changing data distributions at deployment time, without access to any labels. To mitigate error accumulation, researchers have widely adopted the teacher-student framework, though its long-term stability is often taken for granted. In this work, we challenge the common strategy of setting the teacher weights to an exponential moving average of the student by showing that error accumulation still occurs, although it is mostly apparent on longer sequences compared to those commonly utilized. We analyze the stability-plasticity trade-off within the teacher-student framework and propose to use an intransigent teacher that does not update its weights. Surprisingly, we show that this simple change allows TTA methods to significantly improve their performance on multiple datasets with longer scenarios and result in increased robustness to changes in hyperparameters. Finally, we show that those changes can be seamlessly and effectively applied to various architectures and experimental setups, including semantic segmentation. The code is available at https://github.com/dmn-sjk/intransigent_teacher.
CommentsAccepted to the Conference on Lifelong Learning Agents (CoLLAs) 2026
