Masked Autoencoders (MAE)
  • a self-supervised representation learner that reconstructs missing parts but is not a generator