conftrace_

2022 NeurIPS NeurIPS 2022

Test-Time Training with Masked Autoencoders

Abstract

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.

Authors

Yossi Gandelsman, Yu Sun, Xinlei Chen, Alexei Efros

Topics

Machine Learning > Learning Types > Self-Supervised Learning Machine Learning > Application Areas > Domain Adaptation Deep Learning > Architectures > Autoencoders Deep Learning > Learning Types > Self-Supervised Learning Deep Learning > Learning Types > Domain Adaptation

Keywords

domain adaptation self-supervised learning distribution shift masked autoencoder test-time training

More from NeurIPS 2022

All NeurIPS 2022 papers →

Learning Multi-resolution Functional Maps with Spectral Attention for Robust Shape Matching

Label-invariant Augmentation for Semi-Supervised Graph Classification

Practical Adversarial Multivalid Conformal Prediction

Constraining Gaussian Processes to Systems of Linear Ordinary Differential Equations

So3krates: Equivariant attention for interactions on arbitrary length-scales in molecular systems

HyperDomainNet: Universal Domain Adaptation for Generative Adversarial Networks