Backdoor Attack Classification on HSOL
100ASRRIPPLES
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RIPPLEStraining_mode=Fine-tuned2022.06 | 100 | 94.81 | |
| TrojanLMtraining_mode=Fine-tuned2022.06 | 100 | 95.21 | |
| SOStraining_mode=Fine-tuned2022.06 | 100 | 95.78 | |
| EPtraining_mode=Fine-tuned2022.06 | 100 | 95.25 | |
| VanillaAttack=Addsent, Target LM=BERT-base2024.02 | 100 | 92.27 | |
| PrefixAttack=Addsent, Target LM=BERT-base2024.02 | 100 | 93.8 | |
| RAPAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 100 | 91.29 | |
| No-DefenseAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 99.99 | 91.59 | |
| No-DefenseAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 99.01 | 89.52 | |
| VanillaAttack=Badnets, Target LM=BERT-base2024.02 | 98.39 | 93.24 | |
| No-DefenseAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 98.27 | 90.28 | |
| RAPAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 98.12 | 89.52 | |
| VanillaAttack=HiddenKiller, Target LM=BERT-base2024.02 | 97.66 | 92.13 | |
| LWStraining_mode=Fine-tuned2022.06 | 97.26 | 95.65 | |
| ONIONAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 94.18 | 91.16 | |
| LWPtraining_mode=Fine-tuned2022.06 | 94.15 | 95.82 | |
| LoRAAttack=Addsent, Target LM=BERT-base2024.02 | 93.16 | 90.82 | |
| ONIONAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 90.18 | 89.53 | |
| PrefixAttack=HiddenKiller, Target LM=BERT-base2024.02 | 88.16 | 93.8 | |
| No-DefenseAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 85.47 | 89.08 | |
| RAPAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 84.91 | 88.76 | |
| ONIONAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 80.76 | 87.68 | |
| PrefixAttack=Badnets, Target LM=BERT-base2024.02 | 73.67 | 94.45 | |
| LoRAAttack=HiddenKiller, Target LM=BERT-base2024.02 | 72.22 | 89.58 | |
| VanillaAttack=StyleBkd, Target LM=BERT-base2024.02 | 68.92 | 94.81 | |
| AdapterAttack=Addsent, Target LM=BERT-base2024.02 | 67.31 | 83.62 | |
| BadActsAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 62.77 | 86.37 | |
| RAPAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 59.35 | 71.6 | |
| LoRAAttack=Badnets, Target LM=BERT-base2024.02 | 54.18 | 91.99 | |
| AdapterAttack=HiddenKiller, Target LM=BERT-base2024.02 | 49.92 | 84.55 | |
| LoRAAttack=StyleBkd, Target LM=BERT-base2024.02 | 49.85 | 90.06 | |
| AdapterAttack=Badnets, Target LM=BERT-base2024.02 | 49.6 | 85.8 | |
| AdapterAttack=StyleBkd, Target LM=BERT-base2024.02 | 46.7 | 84.71 | |
| PrefixAttack=StyleBkd, Target LM=BERT-base2024.02 | 43.72 | 93.24 | |
| BadActsAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 39.83 | 88.52 | |
| BadLLM-TGAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 29.14 | 89.54 | |
| MuScleLoRAAttack=Badnets, Target LM=BERT-base2024.02 | 24.31 | 86 | |
| BadLLM-TGAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 18.52 | 90.75 | |
| MuScleLoRAAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 18.19 | 89.42 | |
| BadActsAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 16.94 | 89.09 | |
| MuScleLoRAAttack=HiddenKiller, Target LM=BERT-base2024.02 | 13.45 | 86.84 | |
| MuScleLoRAAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 11.93 | 89.54 | |
| BadLLM-TGAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 11.25 | 89.54 | |
| MuScleLoRAAttack=StyleBkd, Target LM=BERT-base2024.02 | 10.63 | 86.64 | |
| ONIONAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 8.57 | 89.26 | |
| BadLLM-TGAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 7.78 | 89.54 | |
| MuScleLoRAAttack=Addsent, Target LM=BERT-base2024.02 | 2.74 | 85.47 | |
| MuScleLoRAAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 1.15 | 89.54 | |
| MuScleLoRAAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 0.8 | 89.34 | |
| BadActsAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 0.3 | 89.54 |