Test-Time Training with Masked Autoencoders

Yossi Gandelsman · Yu Sun · Xinlei Chen · Alexei Efros

Hall J #915

Keywords: [ Test-Time Training ] [ Computer Vision ] [ Masked Auto-Encoder ]


Test-time training adapts to a new test distribution on the fly by optimizing a model for each test input using self-supervision.In this paper, we use masked autoencoders for this one-sample learning problem.Empirically, our simple method improves generalization on many visual benchmarks for distribution shifts.Theoretically, we characterize this improvement in terms of the bias-variance trade-off.

Chat is not available.