“Test-Time Training With Masked Autoencoders”, 2022-09-15 (; similar):
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.