MT3: Meta Test-Time Training for Self-Supervised Test-Time Adaption

Alexander Bartler, Andre Bühler, Felix Wiewel, Mario Döbler, Bin Yang
Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, PMLR 151:3080-3090, 2022.

Abstract

An unresolved problem in Deep Learning is the ability of neural networks to cope with domain shifts during test-time, imposed by commonly fixing network parameters after training. Our proposed method Meta Test-Time Training (MT3), however, breaks this paradigm and enables adaption at test-time. We combine meta-learning, self-supervision and test-time training to learn to adapt to unseen test distributions. By minimizing the self-supervised loss, we learn task-specific model parameters for different tasks. A meta-model is optimized such that its adaption to the different task-specific models leads to higher performance on those tasks. During test-time a single unlabeled image is sufficient to adapt the meta-model parameters. This is achieved by minimizing only the self-supervised loss component resulting in a better prediction for that image. Our approach significantly improves the state-of-the-art results on the CIFAR-10-Corrupted image classification benchmark.

Cite this Paper

BibTeX
@InProceedings{pmlr-v151-bartler22a, title = { MT3: Meta Test-Time Training for Self-Supervised Test-Time Adaption }, author = {Bartler, Alexander and B\"uhler, Andre and Wiewel, Felix and D\"obler, Mario and Yang, Bin}, booktitle = {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics}, pages = {3080--3090}, year = {2022}, editor = {Camps-Valls, Gustau and Ruiz, Francisco J. R. and Valera, Isabel}, volume = {151}, series = {Proceedings of Machine Learning Research}, month = {28--30 Mar}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v151/bartler22a/bartler22a.pdf}, url = {https://proceedings.mlr.press/v151/bartler22a.html}, abstract = { An unresolved problem in Deep Learning is the ability of neural networks to cope with domain shifts during test-time, imposed by commonly fixing network parameters after training. Our proposed method Meta Test-Time Training (MT3), however, breaks this paradigm and enables adaption at test-time. We combine meta-learning, self-supervision and test-time training to learn to adapt to unseen test distributions. By minimizing the self-supervised loss, we learn task-specific model parameters for different tasks. A meta-model is optimized such that its adaption to the different task-specific models leads to higher performance on those tasks. During test-time a single unlabeled image is sufficient to adapt the meta-model parameters. This is achieved by minimizing only the self-supervised loss component resulting in a better prediction for that image. Our approach significantly improves the state-of-the-art results on the CIFAR-10-Corrupted image classification benchmark. } }
Endnote
%0 Conference Paper %T MT3: Meta Test-Time Training for Self-Supervised Test-Time Adaption %A Alexander Bartler %A Andre Bühler %A Felix Wiewel %A Mario Döbler %A Bin Yang %B Proceedings of The 25th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2022 %E Gustau Camps-Valls %E Francisco J. R. Ruiz %E Isabel Valera %F pmlr-v151-bartler22a %I PMLR %P 3080--3090 %U https://proceedings.mlr.press/v151/bartler22a.html %V 151 %X An unresolved problem in Deep Learning is the ability of neural networks to cope with domain shifts during test-time, imposed by commonly fixing network parameters after training. Our proposed method Meta Test-Time Training (MT3), however, breaks this paradigm and enables adaption at test-time. We combine meta-learning, self-supervision and test-time training to learn to adapt to unseen test distributions. By minimizing the self-supervised loss, we learn task-specific model parameters for different tasks. A meta-model is optimized such that its adaption to the different task-specific models leads to higher performance on those tasks. During test-time a single unlabeled image is sufficient to adapt the meta-model parameters. This is achieved by minimizing only the self-supervised loss component resulting in a better prediction for that image. Our approach significantly improves the state-of-the-art results on the CIFAR-10-Corrupted image classification benchmark.
APA
Bartler, A., Bühler, A., Wiewel, F., Döbler, M. & Yang, B.. (2022). MT3: Meta Test-Time Training for Self-Supervised Test-Time Adaption . Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 151:3080-3090 Available from https://proceedings.mlr.press/v151/bartler22a.html.

Related Material