- TL;DR: Training on a single test input with self-supervision makes the prediction better on this input when it is out-of-distribution.
- Abstract: 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 the test sample. 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.
- Code: https://drive.google.com/open?id=1xw-NylSnEjyHs67TXAptviOsx4YuSuZZ
- Keywords: out-of-distribution, distribution shifts