Text Classification on IMDB (test) (Adversarial Robustness via AUA and Query Counts)
95.3Clean AccuracyFreeLB++
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FreeLB++PLM=RoBERTa-base2024.06 | 95.3 | 46 | 42.1 | 33.9 | 829 | 974 | 637 | |
| DiffuseDef-FreeLB++PLM=RoBERTa-base2024.06 | 95 | 86.2 | 85.9 | 86.8 | 3,573 | 4,663 | 2,941 | |
| InfoBERTPLM=RoBERTa-base2024.06 | 94.8 | 30.9 | 27.9 | 21.8 | 681 | 760 | 549 | |
| Fine-TunedPLM=RoBERTa-base2024.06 | 94.6 | 21.3 | 17.9 | 13.6 | 587 | 671 | 493 | |
| DiffuseDef-FreeLB++Backbone=BERT2024.06 | 94.4 | 82.1 | 83 | 84 | 3,174 | 4,348 | 2,842 | |
| FreeLB++Backbone=BERT2024.06 | 94.3 | 44.2 | 39.6 | 40.6 | 784 | 829 | 426 | |
| ATINTERBackbone=BERT2024.06 | 94.3 | 25 | 15 | 39 | 1,173 | 645 | 637 | |
| InfoBERTBackbone=BERT2024.06 | 93.9 | 29.2 | 25.4 | 30.7 | 642 | 644 | 390 | |
| RMLMBackbone=BERT2024.06 | 93.6 | 48.8 | 47.9 | 47.1 | 2,554 | 3,258 | 2,064 | |
| Fine-TunedBackbone=BERT2024.06 | 93.3 | 7.7 | 8.3 | 10.5 | 540 | 534 | 378 | |
| RSMIPLM=RoBERTa-base2024.06 | 92.7 | 77.9 | 74.3 | 70.6 | 3,443 | 4,342 | 2,619 | |
| EarlyRobustBackbone=BERT2024.06 | 92.7 | 49.7 | 46.8 | 43.8 | 2,267 | 2,788 | 1,841 | |
| DiffuseDef-RSMIPLM=RoBERTa-base2024.06 | 92.4 | 84.7 | 84.1 | 84.3 | 3,673 | 4,782 | 3,007 | |
| RSMIBackbone=BERT2024.06 | 90.9 | 60 | 54.4 | 51.1 | 2,840 | 3,455 | 2,070 | |
| DiffuseDef-RSMIBackbone=BERT2024.06 | 90.2 | 80.9 | 79.8 | 79.8 | 3,590 | 4,748 | 2,901 |