MT3: Meta Test-Time Training for Self-Supervised Test-Time Adaption
arXiv:2103.16201
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. Our implementation is available on GitHub.
References in corpus (12)
- Bootstrap your own latent: A new approach to self-supervised Learning
- Deep Visual Domain Adaptation: A Survey
- Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
- On First-Order Meta-Learning Algorithms
- Do ImageNet Classifiers Generalize to ImageNet?
- Out-of-Distribution Generalization via Risk Extrapolation (REx)
- Unsupervised Learning via Meta-Learning
- One-Shot Unsupervised Cross Domain Translation
- Meta-Learning Update Rules for Unsupervised Representation Learning
- How to train your MAML
- Adversarial Style Mining for One-Shot Unsupervised Domain Adaptation
- Improving out-of-distribution generalization via multi-task self-supervised pretraining