Adversarial Bias Mitigation on HotpotQA
-0.015Unattacked Avg |PS|BiasDef
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BiasDefBackbone=DeepSeek-R1-Distill-Qwen-14B, Injection Type=Negative, Injection Count=12025.11 | -0.015 | 0.33 | |
| BiasDefBackbone=DeepSeek-R1-Distill-Qwen-14B, Injection Type=Negative, Injection Count=52025.11 | -0.015 | 0.6 | |
| BiasDefBackbone=DeepSeek-R1-Distill-Qwen-14B, Injection Type=Negative, Injection Count=102025.11 | -0.015 | 0.4 | |
| BiasDefBackbone=DeepSeek-R1-Distill-Qwen-14B, Injection Type=Positive, Injection Count=12025.11 | -0.015 | 0.13 | |
| BiasDefBackbone=DeepSeek-R1-Distill-Qwen-14B, Injection Type=Positive, Injection Count=52025.11 | -0.015 | 0.73 | |
| BiasDefBackbone=DeepSeek-R1-Distill-Qwen-14B, Injection Type=Positive, Injection Count=102025.11 | -0.015 | 1.07 |