A Probabilistic Framework for Lifelong Test-Time Adaptation
This work addresses the challenge of adapting machine learning models during inference in dynamic, real-world environments where data distributions change continuously, which is crucial for applications like autonomous systems or medical diagnostics, though it builds incrementally on prior TTA methods.
The paper tackles the problem of lifelong test-time adaptation (TTA) where test input distributions shift continually over time, and existing methods lack reliable uncertainty estimates. It introduces PETAL, a probabilistic framework that reduces predictive error rates and improves uncertainty metrics like Brier score and negative log-likelihood, achieving better results than state-of-the-art methods on benchmarks such as CIFAR-10C, CIFAR-100C, ImageNetC, and ImageNet3DCC datasets.
Test-time adaptation (TTA) is the problem of updating a pre-trained source model at inference time given test input(s) from a different target domain. Most existing TTA approaches assume the setting in which the target domain is stationary, i.e., all the test inputs come from a single target domain. However, in many practical settings, the test input distribution might exhibit a lifelong/continual shift over time. Moreover, existing TTA approaches also lack the ability to provide reliable uncertainty estimates, which is crucial when distribution shifts occur between the source and target domain. To address these issues, we present PETAL (Probabilistic lifElong Test-time Adaptation with seLf-training prior), which solves lifelong TTA using a probabilistic approach, and naturally results in (1) a student-teacher framework, where the teacher model is an exponential moving average of the student model, and (2) regularizing the model updates at inference time using the source model as a regularizer. To prevent model drift in the lifelong/continual TTA setting, we also propose a data-driven parameter restoration technique which contributes to reducing the error accumulation and maintaining the knowledge of recent domains by restoring only the irrelevant parameters. In terms of predictive error rate as well as uncertainty based metrics such as Brier score and negative log-likelihood, our method achieves better results than the current state-of-the-art for online lifelong test-time adaptation across various benchmarks, such as CIFAR-10C, CIFAR-100C, ImageNetC, and ImageNet3DCC datasets. The source code for our approach is accessible at https://github.com/dhanajitb/petal.