List your company
ResearchFormerly WizWand

Natural Language Inference on MultiNLI matched (test)

85.38AccuracyBERT-base + PoE
63.976869.533475.0980.6466Jun 1, 2017Mar 20, 2018Jan 7, 2019Oct 27, 2019Aug 15, 2020Jun 4, 2021Mar 24, 2022
Updated 8mo ago

Evaluation results

MethodLinks
BERT-base + PoEDebiasing Strategy=Z-Aug2022.0385.38
BERT-baseDebiasing Strategy=Z-Aug2022.0385.12
BERT-base + PoETraining Data=D_MNLI2022.0384.69
PoETraining Data=MNLI2022.0384.11
BERT-baseTraining Data=D_MNLI baseline2022.0384.11
Regularized-confTraining Data=MNLI2022.0384.1
BERT-baseDebiasing Strategy=Seq-Z2022.0383.41
BERT-baseDebiasing Strategy=Par-Z2022.0383.27
DRCN+ELMoensemble=true, embedding=ELMo2018.0582.3
LM-Transformerensemble=false2018.0582.1
DRCNensemble=true2018.0580.6
CAFEensemble=true2018.0580.2
DIINEnsemble=true2017.0980
DIINensemble=true2018.0580
Learned-MixinTraining Data=MNLI2022.0379.5
DRCNensemble=false2018.0579.1
DIINEnsemble=false2017.0978.8
DIINensemble=false2018.0578.8
CAFEensemble=false2018.0578.7
CAFE2017.1178.7
DIINExtra SNLI training set=true2017.1177.8
KIM2017.1177.2
ESIM2017.1176.8
Gated-Att BiLSTMEnsemble=true2017.0974.9
Shortcut-Stacked encoderEnsemble=false2017.0974.6
Shortcut stacked BiLSTM# Params=140M, Additional training data=SNLI dataset2018.0974.6
2-layer Bi-CAS-LSTM# Params=6.8M2018.0974.6
SS BILSTM2017.1174.6
ESIM GloVeAuxiliary Information=GloVe vectors2017.0674.58
Shortcut-Stacked BiLSTM encodersSNLI Mix=true, Parameter count=140.2m2017.1274.5
Shortcut-Stacked BiLSTM2018.0874.5
3-layer Bi-CAS-LSTM# Params=8.6M2018.0974.2
Distance-based Self-Attention NetworkSNLI Mix=false, Parameter count=4.7m2017.1274.1
2-layer CAS-LSTM# Params=2.9M2018.0974
BiLSTM with generalized pooling# Params=18M2018.0973.8
3-layer CAS-LSTM# Params=4.8M2018.0973.8
HBMPdimensions=600D2018.0873.7
Deep Gated Attn. BiLSTM encodersSNLI Mix=false, Parameter count=11.6m2017.1273.5
Deep Gated Attn. BiLSTM encoders2018.0873.5
BiLSTM with gated pooling# Params=12M2018.0973.5
Gated BiLSTM2017.1173.5
Gated-Att BiLSTMEnsemble=false2017.0973.2
ESIMEnsemble=false2017.0972.3
ESIMensemble=false2018.0572.3
InnerAttEnsemble=true2017.0972.2
BiLSTM + Inner-attentionSNLI Mix=true2017.1272.1
InnerAttEnsemble=false2017.0972.1
BiLSTM + Inner-attention2018.0872.1
ESIM dictionaryAuxiliary Information=dictionary definitions2017.0671.45
Directional Self-Attention NetworkSNLI Mix=false, Parameter count=2.4m2017.1271
DiSAN2017.1171
BiLSTM + enhanced embedding + max poolingSNLI Mix=false2017.1270.7
BiLSTM + enh embed + max pooling2018.0870.7
ESIM spellingAuxiliary Information=spelling2017.0668.89
ESIM baselineAuxiliary Information=none2017.0668.55
Cha-level Intra-attention BiLSTM encodersSNLI Mix=true2017.1267.9
BiLSTMSNLI Mix=true2017.1267.5
BiLSTM2018.0867.5
BiLSTMEnsemble=false2017.0967
BiLSTM# Params=null2018.0966.9
BILSTM2017.1166.9
CBOWSNLI Mix=true2017.1266.2
CBOW2018.0866.2
CBOW# Params=null2018.0964.8
CBOW2017.1164.8