search

EdinburghNLP / nematus

Open-Source Neural Machine Translation in Tensorflow
BSD 3-Clause "New" or "Revised" License
797 stars 269 forks source link

Reward Augmented Maximum Likelihood #52

Closed Proyag closed 7 years ago

Proyag commented 7 years ago

Added the Reward Augmented Maximum Likelihood (RAML) training objective (https://arxiv.org/pdf/1609.00150.pdf), which has shown better results than CE training in basic experiments so far.