The AFRL WMT17 Neural Machine Translation Training Task Submission
暂无分享,去创建一个
The WMT17 Neural Machine Translation Training Task aims to test various methods of training neural machine translation systems. We describe the AFRL submission, including preprocessing and its knowledge distillation framework. Teacher systems are given factors for domain, case, and subword location. Student systems are given multiple teachers’ output and a subselected set of the training data designed to match the target domain. Numerical results indicate that the student systems surpass the teachers in translation quality and that this benefit comes directly from the inclusion of the teachers’ output.
[1] Rich Caruana,et al. Do Deep Nets Really Need to be Deep? , 2013, NIPS.
[2] Elizabeth Salesky,et al. The AFRL-MITLL WMT16 News-Translation Task Systems , 2016, WMT.
[3] Philip Gage,et al. A new algorithm for data compression , 1994 .