Test-Time Training with Self-Supervision for Generalization under Distribution Shifts

Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A. Efros +1 more
9/29/2019
cs.LGcs.CVstat.ML

Official code: yueatsprograms/ttt_cifar_release (161 stars, last commit 2020-08-16). 1 more implementation below.

Abstract

In this paper, we propose Test-Time Training, a general approach for improving the performance of predictive models when training and test data come from different distributions. We turn a single unlabeled test sample into a self-supervised learning problem, on which we update the model parameters before making a prediction. This also extends naturally to data in an online stream. Our simple approach leads to improvements on diverse image classification benchmarks aimed at evaluating robustness to distribution shifts.

View on arXivView PDF

Code Implementations(2)

yueatsprograms/ttt_cifar_releaseOfficial100%

TTT Code Release

16132Shell, PythonSep 29, 20196 years ago
yueatsprograms/ttt_imagenet_releaseOfficial100%

TTT Code Release

12613Shell, PythonSep 29, 20196 years ago

Similar Papers

Improving Out-of-Distribution Robustness via Selective Augmentation

85% match1 repo522022

Test-Time Learning for Large Language Models

85% match2025

Free on the Fly: Enhancing Flexibility in Test-Time Adaptation with Online EM

85% match2025

Probabilistic Self-supervised Learning via Scoring Rules Minimization

85% match2023

TIPI: Test Time Adaptation with Transformation Invariance

85% match2023

CTA: Cross-Task Alignment for Better Test Time Training

85% match2025

Cite this paper

@article{sun2019testtime,
  title  = {Test-Time Training with Self-Supervision for Generalization under Distribution Shifts},
  author = {Yu Sun and Xiaolong Wang and Zhuang Liu and John Miller and Alexei A. Efros and Moritz Hardt},
  year   = {2019},
  eprint = {1909.13231},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1909.13231}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1909.13231.svg)](https://codeofpaper.com/paper/1909.13231)

Discussion