M2T: Masking Transformers Twice for Faster Decoding | IEEE Conference Publication | IEEE Xplore