TTAPS: Test-Time Adaption by Aligning Prototypes using Self-Supervision
This addresses the issue of model robustness to distribution shifts for users of deep learning systems, but it is incremental as it modifies an existing self-supervised method.
The paper tackled the problem of performance degradation in deep neural networks due to input distribution drift by proposing a test-time adaptation method that aligns representations with self-supervised prototypes, achieving success on the CIFAR10-C benchmark dataset.
Nowadays, deep neural networks outperform humans in many tasks. However, if the input distribution drifts away from the one used in training, their performance drops significantly. Recently published research has shown that adapting the model parameters to the test sample can mitigate this performance degradation. In this paper, we therefore propose a novel modification of the self-supervised training algorithm SwAV that adds the ability to adapt to single test samples. Using the provided prototypes of SwAV and our derived test-time loss, we align the representation of unseen test samples with the self-supervised learned prototypes. We show the success of our method on the common benchmark dataset CIFAR10-C.