TTAPS: Test-Time Adaption by Aligning Prototypes using Self-Supervision

TL;DR

This paper introduces TTAPS, a test-time adaptation method that aligns test sample representations with learned prototypes using self-supervision, improving robustness against distribution shifts.

Contribution

It proposes a novel test-time adaptation technique based on SwAV that aligns test samples with prototypes, enhancing model performance under distribution shifts.

Findings

Effective on CIFAR10-C benchmark

Improves robustness to distribution shifts

Utilizes self-supervised prototypes for adaptation

Abstract

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.