MAR: Masked Autoencoders for Efficient Action Recognition | Read Paper on Bytez