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

TL;DR

This paper introduces Test-Time Training, a method that enhances model robustness to distribution shifts by self-supervised adaptation on individual test samples, improving performance on image classification tasks.

Contribution

It presents a novel test-time training approach that uses self-supervision on single test samples to adapt models for better generalization under distribution shifts.

Findings

Improves accuracy on distribution-shifted image classification benchmarks

Effective in online streaming scenarios

Simple and generalizable approach

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.