Fairness and Utility Evaluation on BBQ, UnQover, CEB Suite, CrowS, ARC-C, and GSM8K Aggregate
97.6BBQ AccuracySelf-Debias Iter2 + Revise
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Self-Debias Iter2 + ReviseCategory=Inference-time Correction, Base Model=Self-Debias Iter22026.04 | 97.6 | 99.2 | 65.3 | 62.5 | 72.3 | 66.8 | 93.7 | 87.7 | 80.6 | |
| Self-Debias Iter2 + ConfirmationCategory=Inference-time Correction, Base Model=Self-Debias Iter22026.04 | 97.2 | 99.3 | 66.4 | 63.6 | 72 | 67.5 | 93.4 | 88.8 | 81 | |
| Self-Debias Iter2 + Self-refineCategory=Inference-time Correction, Base Model=Self-Debias Iter22026.04 | 97.1 | 99.4 | 68.1 | 63.3 | 72.5 | 68.8 | 93.6 | 89.5 | 81.5 | |
| Self-Debias Iter2 + Self-CorrectionCategory=Debiasing Methods (Ours), Base Model=Self-Debias Iter22026.04 | 97.1 | 99.5 | 68.1 | 65.8 | 71.9 | 72.2 | 93 | 89.5 | 82.1 | |
| Self-Debias Iter2Iteration=22026.04 | 97 | 99.5 | 67.1 | 65.8 | 72.1 | 71.2 | 93.1 | 87.6 | 81.7 | |
| Self-Debias Iter2 + DenyingCategory=Inference-time Correction, Base Model=Self-Debias Iter22026.04 | 96.7 | 99.4 | 66.9 | 62.3 | 71.5 | 64.8 | 92.9 | 88.6 | 80.4 | |
| Self-Debias Iter2 + CALCategory=Debiasing Methods, Base Model=Self-Debias Iter22026.04 | 96.5 | 99.7 | 64.7 | 63.9 | 72.8 | 66.8 | 93 | 87.3 | 80.6 | |
| Self-Debias Iter2 + BiasFilterCategory=Debiasing Methods, Base Model=Self-Debias Iter22026.04 | 95.7 | 99.6 | 54.5 | 64.3 | 70.8 | 65.2 | 92.6 | 86.4 | 78.6 |