Backdoor Attack Defense on Backdoor Attacks (test)
14.7ASRCLEANGEN
Evaluation Results
| Method | Links | |
|---|---|---|
| CLEANGENType=Reactive, Attacked Model=Qwen-2.5-7B2026.05 | 14.7 | |
| OBBRType=Proactive, Attacked Model=Qwen-2.5-7B2026.05 | 16.5 | |
| DPRType=Proactive, Attacked Model=Qwen-2.5-7B2026.05 | 28.4 | |
| CBBRType=Proactive, Attacked Model=Qwen-2.5-7B2026.05 | 28.5 | |
| ParaphraseType=Proactive, Attacked Model=Qwen-2.5-7B2026.05 | 29.1 | |
| OBBRType=Proactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 30.4 | |
| OBBRType=Proactive, Attacked Model=Avg.2026.05 | 30.7 | |
| OBBRType=Proactive, Attacked Model=Llama-3.2-1B2026.05 | 31.2 | |
| DPRType=Proactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 34.7 | |
| CBBRType=Proactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 35.6 | |
| ParaphraseType=Proactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 37.7 | |
| CBBRType=Proactive, Attacked Model=Avg.2026.05 | 40.2 | |
| ParaphraseType=Proactive, Attacked Model=Avg.2026.05 | 41 | |
| DPRType=Proactive, Attacked Model=Avg.2026.05 | 42.9 | |
| OBBRType=Proactive, Attacked Model=Llama-3.1-8B2026.05 | 44.6 | |
| CBBRType=Proactive, Attacked Model=Llama-3.2-1B2026.05 | 45.9 | |
| ParaphraseType=Proactive, Attacked Model=Llama-3.1-8B2026.05 | 47 | |
| CLEANGENType=Reactive, Attacked Model=Avg.2026.05 | 49 | |
| ParaphraseType=Proactive, Attacked Model=Llama-3.2-1B2026.05 | 50.3 | |
| CBBRType=Proactive, Attacked Model=Llama-3.1-8B2026.05 | 50.8 | |
| DecodingType=Reactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 53.3 | |
| DPRType=Proactive, Attacked Model=Llama-3.2-1B2026.05 | 53.9 | |
| DPRType=Proactive, Attacked Model=Llama-3.1-8B2026.05 | 54.4 | |
| CLEANGENType=Reactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 56 | |
| QuantizeType=Reactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 58.8 | |
| CLEANGENType=Reactive, Attacked Model=Llama-3.2-1B2026.05 | 59.6 | |
| DecodingType=Reactive, Attacked Model=Qwen-2.5-7B2026.05 | 60.7 | |
| CROWType=Intraactive, Attacked Model=Qwen-2.5-1.5B2026.05 | 63.2 | |
| CLEANGENType=Reactive, Attacked Model=Llama-3.1-8B2026.05 | 65.5 | |
| CROWType=Intraactive, Attacked Model=Llama-3.1-8B2026.05 | 66.3 | |
| DecodingType=Reactive, Attacked Model=Avg.2026.05 | 67.1 | |
| CROWType=Intraactive, Attacked Model=Avg.2026.05 | 68.6 | |
| BaseType=None, Attacked Model=Qwen-2.5-7B2026.05 | 68.7 | |
| CROWType=Intraactive, Attacked Model=Llama-3.2-1B2026.05 | 68.8 | |
| BaseType=None, Attacked Model=Qwen-2.5-1.5B2026.05 | 69.1 | |
| QuantizeType=Reactive, Attacked Model=Qwen-2.5-7B2026.05 | 70.7 | |
| QuantizeType=Reactive, Attacked Model=Avg.2026.05 | 71.5 | |
| DecodingType=Reactive, Attacked Model=Llama-3.2-1B2026.05 | 72.6 | |
| BaseType=None, Attacked Model=Avg.2026.05 | 74.6 | |
| QuantizeType=Reactive, Attacked Model=Llama-3.2-1B2026.05 | 76 | |
| CROWType=Intraactive, Attacked Model=Qwen-2.5-7B2026.05 | 76.1 | |
| BaseType=None, Attacked Model=Llama-3.2-1B2026.05 | 76.5 | |
| QuantizeType=Reactive, Attacked Model=Llama-3.1-8B2026.05 | 80.4 | |
| DecodingType=Reactive, Attacked Model=Llama-3.1-8B2026.05 | 81.9 | |
| BaseType=None, Attacked Model=Llama-3.1-8B2026.05 | 84.2 |