Machine Reading Comprehension on RACE Full (test)
53.3AccuracyBiAttention (MRU) Ensemble
Evaluation results
| Method | Links | |
|---|---|---|
| BiAttention (MRU) EnsembleEnsemble size=92018.03 | 53.3 | |
| DFN EnsembleEnsemble size=92018.03 | 51.2 | |
| BiAttention (Sim. MRU)Encoder architecture=Sim. MRU, Hidden dimension=250d, Training time=4 min (12 hours)2018.03 | 50.4 | |
| BiAttention (MRU)Encoder architecture=MRU, Hidden dimension=250d, Training time=12 min (20 hours)2018.03 | 50 | |
| Dynamic Fusion NetworkTraining time=≈8 hours (1 week*)2018.03 | 47.4 | |
| GA + ElimiNet2018.03 | 47.2 | |
| BiAttentionEncoder architecture=No Encoder, Training time=3 min (9 hours)2018.03 | 44.9 | |
| ElimiNet2018.03 | 44.5 | |
| GA2018.03 | 44.1 | |
| BiAttentionEncoder architecture=GRU, Hidden dimension=250d, Training time=16 min (2 days)2018.03 | 44 | |
| BiAttentionEncoder architecture=LSTM, Hidden dimension=250d, Training time=18 min (2 days)2018.03 | 43.6 | |
| Stanford AR2018.03 | 43.3 | |
| Sliding Window2018.03 | 32.2 |