Direct synthesis prediction on USPTO-MIT separated
92Top-1 AccuracyAugmented Transformer
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Augmented TransformerAugmentation=x100, Beam size=10, Training set size=400k2020.03 | 92 | 95.4 | 97 | |
| Augmented TransformerAugmentation=x100, Beam size=10, Training set size=439k2020.03 | 91.9 | 95.4 | 97 | |
| Transformer (ensemble)Ensemble=true2020.03 | 91 | 94.3 | 95.8 | |
| Transformer (single)Ensemble=false2020.03 | 90.4 | 93.7 | 95.3 | |
| WLDN52020.03 | 85.6 | — | 93.4 | |
| GTPN2020.03 | 83.2 | — | 86.5 | |
| Seq2Seq2020.03 | 80.3 | — | 87.5 | |
| WLDN2020.03 | 79.6 | — | 89.2 |