Text Classification on AGNews (test) (Adversarial Robustness & Query Efficiency)
95.5Accuracy (Clean)InfoBERT
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| InfoBERTPLM=RoBERTa-base2024.06 | 95.5 | 40.2 | 45.2 | 48.6 | 392 | 421 | 430 | |
| FreeLB++PLM=RoBERTa-base2024.06 | 95.4 | 57.5 | 62.9 | 55.9 | 444 | 467 | 447 | |
| DiffuseDef-FreeLB++PLM=RoBERTa-base2024.06 | 95.3 | 85.6 | 87.6 | 85.3 | 880 | 976 | 906 | |
| InfoBERTBackbone=BERT2024.06 | 95 | 35.5 | 39.1 | 42.6 | 377 | 397 | 397 | |
| FreeLB++Backbone=BERT2024.06 | 95 | 54.7 | 56.5 | 44.6 | 426 | 430 | 390 | |
| Fine-TunedPLM=RoBERTa-base2024.06 | 94.9 | 34.1 | 36.9 | 43.6 | 372 | 396 | 410 | |
| DiffuseDef-FreeLB++Backbone=BERT2024.06 | 94.8 | 84.5 | 86 | 84.6 | 877 | 972 | 910 | |
| Fine-TunedBackbone=BERT2024.06 | 94.4 | 10.2 | 25.4 | 27.1 | 348 | 372 | 379 | |
| EarlyRobustBackbone=BERT2024.06 | 94.4 | 35.6 | 37.2 | 45.7 | 475 | 516 | 533 | |
| RMLMBackbone=BERT2024.06 | 94.3 | 45.6 | 44.4 | 54.1 | 642 | 738 | 705 | |
| RSMIBackbone=BERT2024.06 | 94.3 | 52.6 | 56.7 | 55.4 | 680 | 737 | 687 | |
| ATINTERBackbone=BERT2024.06 | 94.2 | 68 | 59 | 81 | 527 | 235 | 122 | |
| DiffuseDef-RSMIBackbone=BERT2024.06 | 93.8 | 82.7 | 83.3 | 84.4 | 894 | 1,029 | 930 | |
| RSMIPLM=RoBERTa-base2024.06 | 93.1 | 64.2 | 66.4 | 67.4 | 774 | 861 | 808 | |
| DiffuseDef-RSMIPLM=RoBERTa-base2024.06 | 92.9 | 82.9 | 83.5 | 82.2 | 905 | 925 | 1,047 |