Bias Defense on Average of four datasets (test)
92.38AccuracyNo Defense (Before)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No Defense (Before)Backbone=RoBERTa, Attack Type=Bias Word2023.05 | 92.38 | 43.75 | |
| Before (No Defense)Backbone=ROBERTa, Attack=BiasWord2023.05 | 92.38 | 43.75 | |
| No Defense (Before)Backbone=RoBERTa, Attack Type=Average2023.05 | 92.26 | 43.75 | |
| NADBackbone=RoBERTa, Attack Type=Bias Word2023.05 | 92.17 | 43.91 | |
| No Defense (Before)Backbone=RoBERTa, Attack Type=BiasSent2023.05 | 92.14 | 43.75 | |
| Before (No Defense)Backbone=ROBERTa, Attack=BiasSent2023.05 | 92.14 | 43.75 | |
| NADBackbone=RoBERTa, Attack Type=Average2023.05 | 92.13 | 43.84 | |
| NADBackbone=RoBERTa, Attack Type=BiasSent2023.05 | 92.08 | 43.78 | |
| KDBackbone=RoBERTa, Attack Type=Bias Word2023.05 | 92.01 | 43.75 | |
| KDBackbone=RoBERTa, Attack Type=Average2023.05 | 91.97 | 43.75 | |
| KDBackbone=RoBERTa, Attack Type=BiasSent2023.05 | 91.93 | 43.75 | |
| NADBackbone=BERT, Attack Type=BiasSent2023.05 | 91.66 | 44.65 | |
| Fine-tuningBackbone=ROBERTa, Attack=BiasSent2023.05 | 91.6 | 44.06 | |
| No Defense (Before)Backbone=BERT, Attack Type=BiasSent2023.05 | 91.44 | 43.75 | |
| Before (No Defense)Backbone=BERT, Attack=BiasSent2023.05 | 91.44 | 43.75 | |
| NADBackbone=BERT, Attack Type=Average2023.05 | 91.42 | 44.73 | |
| No Defense (Before)Backbone=BERT, Attack Type=Average2023.05 | 91.35 | 43.75 | |
| Fine-tuningBackbone=ROBERTa, Attack=BiasWord2023.05 | 91.3 | 43.75 | |
| No Defense (Before)Backbone=BERT, Attack Type=Bias Word2023.05 | 91.27 | 43.75 | |
| Before (No Defense)Backbone=BERT, Attack=BiasWord2023.05 | 91.27 | 43.75 | |
| NADBackbone=BERT, Attack Type=Bias Word2023.05 | 91.18 | 44.82 | |
| KDBackbone=BERT, Attack Type=BiasSent2023.05 | 91.03 | 43.75 | |
| Fine-tuningBackbone=BERT, Attack=BiasWord2023.05 | 90.84 | 43.75 | |
| Fine-tuningBackbone=BERT, Attack=BiasSent2023.05 | 90.83 | 43.75 | |
| KDBackbone=BERT, Attack Type=Average2023.05 | 90.8 | 43.76 | |
| KDBackbone=BERT, Attack Type=Bias Word2023.05 | 90.57 | 43.76 | |
| RAPBackbone=RoBERTa, Attack Type=Bias Word2023.05 | 89.98 | 88.73 | |
| RAPBackbone=RoBERTa, Attack Type=Average2023.05 | 89.55 | 71.67 | |
| RAPBackbone=BERT, Attack Type=BiasSent2023.05 | 89.4 | 66.47 | |
| RAPBackbone=RoBERTa, Attack Type=BiasSent2023.05 | 89.13 | 54.61 | |
| RAPBackbone=BERT, Attack Type=Average2023.05 | 89.09 | 77.58 | |
| RAPBackbone=BERT, Attack Type=Bias Word2023.05 | 88.79 | 88.69 | |
| ONIONBackbone=BERT, Attack Type=BiasSent2023.05 | 87.82 | 58.65 | |
| ONIONBackbone=BERT, Attack Type=Average2023.05 | 87.5 | 66.89 | |
| ONIONBackbone=BERT, Attack Type=Bias Word2023.05 | 87.12 | 75.14 | |
| Fine-pruningBackbone=ROBERTa, Attack=BiasWord2023.05 | 87.09 | 64.65 | |
| Fine-purifyingBackbone=RoBERTa, Attack Type=BiasSent2023.05 | 86.71 | 84.11 | |
| Fine-purifyingBackbone=ROBERTa, Attack=BiasSent2023.05 | 86.71 | 84.11 | |
| Fine-pruningBackbone=ROBERTa, Attack=BiasSent2023.05 | 86.69 | 76.43 | |
| Fine-purifyingBackbone=RoBERTa, Attack Type=Average2023.05 | 86.56 | 85.2 | |
| ONIONBackbone=RoBERTa, Attack Type=Bias Word2023.05 | 86.42 | 76.8 | |
| Fine-purifyingBackbone=RoBERTa, Attack Type=Bias Word2023.05 | 86.42 | 86.3 | |
| Fine-purifyingBackbone=ROBERTa, Attack=BiasWord2023.05 | 86.42 | 86.3 | |
| ONIONBackbone=RoBERTa, Attack Type=Average2023.05 | 86.4 | 63.53 | |
| ONIONBackbone=RoBERTa, Attack Type=BiasSent2023.05 | 86.37 | 50.26 | |
| Fine-pruningBackbone=BERT, Attack=BiasWord2023.05 | 86.05 | 61.57 | |
| Fine-mixingBackbone=ROBERTa, Attack=BiasSent2023.05 | 86.02 | 77.73 | |
| Fine-mixingBackbone=ROBERTa, Attack=BiasWord2023.05 | 85.92 | 81.79 | |
| Fine-purifyingBackbone=BERT, Attack Type=BiasSent2023.05 | 85.63 | 84.03 | |
| Fine-purifyingBackbone=BERT, Attack=BiasSent2023.05 | 85.63 | 84.03 | |
| Fine-purifyingBackbone=BERT, Attack Type=Average2023.05 | 85.5 | 84.55 | |
| Fine-pruningBackbone=BERT, Attack=BiasSent2023.05 | 85.48 | 64.38 | |
| Fine-purifyingBackbone=BERT, Attack Type=Bias Word2023.05 | 85.38 | 85.06 | |
| Fine-purifyingBackbone=BERT, Attack=BiasWord2023.05 | 85.38 | 85.06 | |
| Fine-mixingBackbone=BERT, Attack=BiasSent2023.05 | 84.81 | 75.26 | |
| Fine-mixingBackbone=BERT, Attack=BiasWord2023.05 | 84.72 | 76.45 |