Backdoor Defense on SST-2
91.71CACCVanilla
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VanillaAttack=StyleBkd, Target LM=BERT-base2024.02 | 91.71 | 77.19 | — | — | — | |
| VanillaTarget LM=BERT-Base, Attack=StyleBkd2024.02 | 91.71 | 77.19 | — | — | — | |
| RAPTarget LM=BERT-Base, Attack=Addsent2024.02 | 91.71 | 27.19 | — | — | — | |
| No-DefenseAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 91.58 | 99.23 | — | — | — | |
| No-DefenseAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 91.49 | 95.92 | — | — | — | |
| No-DefenseAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 91.42 | 70.11 | — | — | — | |
| STRIPTarget LM=BERT-Base, Attack=Addsent2024.02 | 91.39 | 28.62 | — | — | — | |
| VanillaAttack=Badnets, Target LM=BERT-base2024.02 | 91.27 | 94.63 | — | — | — | |
| VanillaAttack=HiddenKiller, Target LM=BERT-base2024.02 | 91.1 | 93.53 | — | — | — | |
| VanillaTarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 91.1 | 93.53 | — | — | — | |
| BadLLM-TGAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 91.09 | 14.76 | — | — | — | |
| VanillaAttack=Addsent, Target LM=BERT-base2024.02 | 90.99 | 99.89 | — | — | — | |
| VanillaTarget LM=BERT-Base, Attack=Addsent2024.02 | 90.99 | 99.89 | — | — | — | |
| CUBETarget LM=BERT-Base, Attack=StyleBkd2024.02 | 90.83 | 22.43 | — | — | — | |
| RAPAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 90.81 | 73.95 | — | — | — | |
| BKITarget LM=BERT-Base, Attack=Addsent2024.02 | 90.72 | 33.05 | — | — | — | |
| No-DefenseAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 90.55 | 94.56 | — | — | — | |
| STRIPTarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 90.39 | 90.57 | — | — | — | |
| BKITarget LM=BERT-Base, Attack=StyleBkd2024.02 | 90.34 | 82.46 | — | — | — | |
| PrefixAttack=Badnets, Target LM=BERT-base2024.02 | 90.17 | 86.73 | — | — | — | |
| RAPTarget LM=BERT-Base, Attack=StyleBkd2024.02 | 90.17 | 79.38 | — | — | — | |
| PrefixAttack=StyleBkd, Target LM=BERT-base2024.02 | 90.06 | 63.6 | — | — | — | |
| RAPAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 90.04 | 98.44 | — | — | — | |
| STRIPTarget LM=BERT-Base, Attack=StyleBkd2024.02 | 89.89 | 78.62 | — | — | — | |
| BadActsAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 89.6 | 50.55 | — | — | — | |
| PrefixAttack=Addsent, Target LM=BERT-base2024.02 | 89.46 | 99.89 | — | — | — | |
| BadActsAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 89.4 | 18.38 | — | — | — | |
| BadActsAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 89.35 | 56.84 | — | — | — | |
| RAPAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 89.31 | 67.21 | — | — | — | |
| RAPAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 89.14 | 94.43 | — | — | — | |
| LoRAAttack=Badnets, Target LM=BERT-base2024.02 | 89.07 | 65.57 | — | — | — | |
| LoRAAttack=Addsent, Target LM=BERT-base2024.02 | 88.96 | 96.16 | — | — | — | |
| BadLLM-TGAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 88.96 | 25.13 | — | — | — | |
| LoRAAttack=StyleBkd, Target LM=BERT-base2024.02 | 88.91 | 57.24 | — | — | — | |
| LoRAAttack=HiddenKiller, Target LM=BERT-base2024.02 | 88.58 | 52.96 | — | — | — | |
| PrefixAttack=HiddenKiller, Target LM=BERT-base2024.02 | 88.52 | 68.64 | — | — | — | |
| BKITarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 88.41 | 94.85 | — | — | — | |
| RAPTarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 88.25 | 89.14 | — | — | — | |
| BadLLM-TGAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 88.24 | 0.86 | — | — | — | |
| BadActsAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 88.19 | 82.3 | — | — | — | |
| MuScleLoRAAttack=StyleBkd, Target LM=BERT-base2024.02 | 87.81 | 33.22 | — | — | — | |
| MuScleLoRATarget LM=BERT-Base, Attack=StyleBkd2024.02 | 87.81 | 33.22 | — | — | — | |
| ONIONAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 87.81 | 86.75 | — | — | — | |
| CUBETarget LM=BERT-Base, Attack=Addsent2024.02 | 87.7 | 37.94 | — | — | — | |
| MuScleLoRAAttack=HiddenKiller, Target LM=BERT-base2024.02 | 87.64 | 25.11 | — | — | — | |
| MuScleLoRATarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 87.64 | 25.11 | — | — | — | |
| ONIONAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 87.37 | 71.09 | — | — | — | |
| ONIONAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 87.28 | 19.43 | — | — | — | |
| AdapterAttack=Badnets, Target LM=BERT-base2024.02 | 87.26 | 55.26 | — | — | — | |
| AdapterAttack=StyleBkd, Target LM=BERT-base2024.02 | 87.1 | 60.96 | — | — | — | |
| ONIONTarget LM=BERT-Base, Attack=Addsent2024.02 | 87.04 | 49.78 | — | — | — | |
| AdapterAttack=Addsent, Target LM=BERT-base2024.02 | 86.88 | 87.83 | — | — | — | |
| MuScleLoRAAttack=Addsent, Target LM=BERT-base2024.02 | 86.77 | 18.97 | — | — | — | |
| MuScleLoRATarget LM=BERT-Base, Attack=Addsent2024.02 | 86.77 | 18.97 | — | — | — | |
| AdapterAttack=HiddenKiller, Target LM=BERT-base2024.02 | 86.6 | 45.39 | — | — | — | |
| MuScleLoRAAttack=Badnets, Target LM=BERT-base2024.02 | 86.54 | 12.94 | — | — | — | |
| CUBETarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 85.5 | 45.61 | — | — | — | |
| ONIONTarget LM=BERT-Base, Attack=StyleBkd2024.02 | 85.45 | 81.76 | — | — | — | |
| ONIONAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 85.36 | 94.78 | — | — | — | |
| ONIONTarget LM=BERT-Base, Attack=HiddenKiller2024.02 | 85.23 | 96.05 | — | — | — | |
| MuScleLoRAAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 84.4 | 30.61 | — | — | — | |
| MuScleLoRAAttack=Sentbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 84.31 | 34.5 | — | — | — | |
| MuScleLoRAAttack=Stylebkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 83.71 | 37.11 | — | — | — | |
| MuScleLoRAAttack=Wordbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 83.54 | 17.96 | — | — | — | |
| BadLLM-TGAttack=Synbkd, Victim Model=BERT-base-uncased, Poisoning Rate=20%2026.03 | 70.26 | 14.36 | — | — | — | |
| MDPAttack=BadNets2023.09 | — | — | — | — | 0.99 | |
| MDPAttack=AddSent2023.09 | — | — | — | — | 0.99 | |
| MDPAttack=LWP2023.09 | — | — | — | — | 0.98 | |
| MDPAttack=EP2023.09 | — | — | — | — | 1 | |
| MDPAttack=SOS2023.09 | — | — | — | — | 1 | |
| ONIONAttack=BadNets2023.09 | — | — | — | — | 0.64 | |
| ONIONAttack=AddSent2023.09 | — | — | — | — | 0.54 | |
| ONIONAttack=LWP2023.09 | — | — | — | — | 0.72 | |
| ONIONAttack=EP2023.09 | — | — | — | — | 0.67 | |
| ONIONAttack=SOS2023.09 | — | — | — | — | 0.61 | |
| RAPAttack=BadNets2023.09 | — | — | — | — | 0.53 | |
| RAPAttack=AddSent2023.09 | — | — | — | — | 0.52 | |
| RAPAttack=LWP2023.09 | — | — | — | — | 0.83 | |
| RAPAttack=EP2023.09 | — | — | — | — | 0.56 | |
| RAPAttack=SOS2023.09 | — | — | — | — | 0.51 | |
| STRIPAttack=BadNets2023.09 | — | — | — | — | 0.66 | |
| STRIPAttack=AddSent2023.09 | — | — | — | — | 0.51 | |
| STRIPAttack=LWP2023.09 | — | — | — | — | 0.6 | |
| STRIPAttack=EP2023.09 | — | — | — | — | 0.84 | |
| STRIPAttack=SOS2023.09 | — | — | — | — | 0.82 |