Monitoring Risks in Test-Time Adaptation
This addresses the challenge of monitoring model degradation during deployment for practitioners using TTA, though it is incremental as it extends existing monitoring tools.
The paper tackles the problem of detecting when test-time adaptation (TTA) methods degrade to failure by proposing a risk monitoring framework that uses sequential testing with confidence sequences to track performance without labels, demonstrating effectiveness across various datasets and shifts.
Encountering shifted data at test time is a ubiquitous challenge when deploying predictive models. Test-time adaptation (TTA) methods address this issue by continuously adapting a deployed model using only unlabeled test data. While TTA can extend the model's lifespan, it is only a temporary solution. Eventually the model might degrade to the point that it must be taken offline and retrained. To detect such points of ultimate failure, we propose pairing TTA with risk monitoring frameworks that track predictive performance and raise alerts when predefined performance criteria are violated. Specifically, we extend existing monitoring tools based on sequential testing with confidence sequences to accommodate scenarios in which the model is updated at test time and no test labels are available to estimate the performance metrics of interest. Our extensions unlock the application of rigorous statistical risk monitoring to TTA, and we demonstrate the effectiveness of our proposed TTA monitoring framework across a representative set of datasets, distribution shift types, and TTA methods.