List your company
ResearchFormerly WizWand

Machine Translation on WMT Chinese-English 2017 (test)

27.4BLEUDLDN
23.999224.882125.76526.6479Mar 15, 2018Apr 21, 2018May 28, 2018Jul 4, 2018Aug 10, 2018Sep 16, 2018Oct 24, 2018
Updated 8mo ago

Evaluation results

MethodLinks
DLDNBack Translation=true, Method=Dual Learning + Deliberation Nets2018.0327.4
ARJTBack Translation=true, Method=Agreement Regularization + Joint Training2018.0327.38
DLDN2Method=Dual Learning + Deliberation Nets, First decoder reranking=false2018.0327.2
ARJT2Method=Agreement Regularization + Joint Training, dropout=0.12018.0327.19
DLDN4Back Translation=true, Method=Dual Learning, Architecture=Bi-NMT2018.0327.16
ARJT3Method=Agreement Regularization + Joint Training, dropout=0.052018.0327.07
ARJT4Method=Agreement Regularization + Joint Training, dropout=0.012018.0326.98
ARBack Translation=true, Method=Agreement Regularization2018.0326.91
DLDN3Back Translation=true, Method=Dual Learning, Sampling=R2L sampling2018.0326.88
DLBack Translation=true, Method=Dual Learning2018.0326.51
SogouContext=WMT 2017 best result, System type=Ensemble2018.0326.4
BTBack Translation=true2018.0325.57
TRANSFORMER-BIG + Localness + Rel_PosArchitecture=TRANSFORMER-BIG, Localness Modeling=True, Relative Position Encoding=True, # Para.=307.3M2018.1025.28
TRANSFORMER-BIG + LocalnessArchitecture=TRANSFORMER-BIG, Localness Modeling=True, # Para.=307.2M2018.1025.03
TRANSFORMER-BASE + Localness + Rel_PosArchitecture=TRANSFORMER-BASE, Localness Modeling=True, Relative Position Encoding=True, # Para.=108.8M2018.1024.96
TRANSFORMER-BASE + LocalnessArchitecture=TRANSFORMER-BASE, Localness Modeling=True, # Para.=108.7M2018.1024.77
TRANSFORMER-BIG (baseline)Architecture=TRANSFORMER-BIG, # Para.=303.9M2018.1024.56
TRANSFORMER-BASE + Rel_PosArchitecture=TRANSFORMER-BASE, Relative Position Encoding=Shaw et al., 2018, # Para.=108.0M2018.1024.53
BaseArchitecture=Transformer Baseline2018.0324.2
TRANSFORMER-BIGArchitecture=TRANSFORMER-BIG, Source=Hassan et al., 20182018.1024.2
TRANSFORMER-BASE (baseline)Architecture=TRANSFORMER-BASE, # Para.=107.9M2018.1024.13