Toxicity detection on CivilComments-WILDS (val)
92.3Average AccuracyERM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ERMBackbone=BERT2021.04 | 92.3 | 53.6 | |
| ERM (ours)Backbone=BERT2021.04 | 92.1 | 54.1 | |
| FishBackbone=BERT2021.04 | 91.8 | 75.3 | |
| Group DROBackbone=BERT2021.04 | 89.6 | 68.7 | |
| ReweightedBackbone=BERT2021.04 | 89.1 | 67.9 |