Self-Guided Masked Autoencoder

Neural Information Processing Systems 

Masked Autoencoder (MAE) is a self-supervised approach for representation learning, widely applicable to a variety of downstream tasks in computer vision.