Test-Time Training for Out-of-Distribution Generalization

By Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A. Efros, Moritz Hardt. 2019

107 citations recorded by the source corpus.

We introduce a general approach, called test-time training, for improving the performance of predictive models when test and training data come from different distributions. Test-time training turns a single unlabeled test instance into a self-supervised learning problem, on which we update the model parameters before making a prediction on this instance. We show that this simple idea leads to surprising improvements on diverse image classification benchmarks aimed at evaluating robustness to distribution shifts. Theoretical investigations on a convex model reveal helpful intuitions for when we can expect our approach to help.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Representation & Self-Supervised Learning

Citations in the corpus

Break-It-Fix-It: Unsupervised Learning for Program Repair (2021, background): “Within domain adaptation, our task is also related to the setting where unlabeled data in the test domain is available (Sun & Saenko, 2016; Hoffman et al., 2018; Sun et al., 2020) (in our case, real bad code).”
On Robustness and Transferability of Convolutional Neural Networks (2020, positive): “Transductive fine-tuning using self-supervision on the test data improves performance under distribution shift [58].”
Adaptive Risk Minimization: A Meta-Learning Approach for Tackling Group Shift (2020, unknown): “Other prior methods that consider unlabeled test time adaptation typically only handle label shift [54, 41, 64] or use carefully chosen surrogate losses, relying on the correlation between this loss and test time performance [65].”
Improving robustness against common corruptions by covariate shift adaptation (2020, methodology): “More involved methods for DA include self-supervised domain adaptation on single examples [49] and pseudo-labeling French et al.”
Tent: Fully Test-Time Adaptation by Entropy Minimization (2021, methodology): “…domain adaptation (UDA-SS) jointly trains self-supervised rotation and position tasks on source and target to optimize for a shared representation (Sun et al., 2019a), • test-time training (TTT) jointly trains for supervised and self-supervised tasks on source, then keeps training the…”
Evaluating Prediction-Time Batch Normalization for Robustness under Covariate Shift (2020, background): “A common theme in these approaches is that they assume access to a set of unlabeled samples from the test set at training time [Sun et al., 2019], whereas our focus is to apply an intervention at test time, after a model has been trained.”
NAS-DIP: Learning Deep Image Prior with Neural Architecture Search (2020, background): “Training CNNs on testing data has been shown as an effective approach for improving performance [55, 57, 76].”
Learning Improvement Heuristics for Solving Routing Problems (2019, background): “This is reasonable because in general, achieving good out-of-distribution generalization is very hard for machine learning models [41].”
Continuous Adaptation for Interactive Object Segmentation by Learning from Corrections (2019, background): “The concurrent work of [54] uses self-supervision to adapt the feature”
Deep Stable Learning for Out-Of-Distribution Generalization (2021, background): “To address this issue, out-of-distribution (OOD) generalization is proposed for improving the generalization ability of models under distribution shifts [55, 27].”
Vision-Language Navigation With Self-Supervised Auxiliary Reasoning Tasks (2019, background): “It has been demonstrated [34, 35] that self-supervised auxiliary tasks facilitate domain adaptation.”