Action-Conditional Video Prediction with motion-equivariance regularizer
Research project involving using a Siamese network and contrasive loss in order to apply motion-equivariance contraint in the embedded space for more general frame prediction in the Atari domain.