wangqiangneu / MT-PaperReading

Record my paper reading about Machine Translation and other related works.
36 stars 2 forks source link

20-ICML-Non-Autoregressive Machine Translation with Disentangled Context Transformer #69

Open wangqiangneu opened 4 years ago

wangqiangneu commented 4 years ago

简介

对CMLM的改进,两个点: 1. CMLM训练时只能训一部分subset (跟bert一样,从这个角度看,跟#25加速bert训练的目标是一致的); 2. CMLM的解码算法比较笨,启发式且必须解码到规定的最大迭代次数上。

DisCo的做法:

有意思的点

论文信息

总结