Backdoor Defense on CR (CA, ASR)
94.32Clean Accuracy (CA)LoRA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LoRAAttack=BadNet, Backbone=Qwen2.52024.10 | 94.32 | 86.07 | |
| LoRAAttack=InSent, Backbone=LLaMA32024.10 | 94.32 | 99.79 | |
| LoRAAttack=BadNet, Backbone=LLaMA32024.10 | 94.06 | 100 | |
| W2SDefenseAttack=BadNet, Backbone=LLaMA32024.10 | 93.81 | 6.24 | |
| W2SDefenseAttack=BadNet, Backbone=Vicuna2024.10 | 93.55 | 8.32 | |
| PruneAttack=InSent, Backbone=LLaMA32024.10 | 93.55 | 42.62 | |
| Back Tr.Attack=BadNet, Backbone=LLaMA32024.10 | 93.16 | 41.37 | |
| Back Tr.Attack=InSent, Backbone=LLaMA32024.10 | 93.16 | 52.39 | |
| PruneAttack=BadNet, Backbone=LLaMA32024.10 | 93.03 | 39.29 | |
| LoRAAttack=BadNet, Backbone=Vicuna2024.10 | 93.03 | 100 | |
| LoRAAttack=SynAttack, Backbone=Qwen2.52024.10 | 92.9 | 22.29 | |
| Back Tr.Attack=InSent, Backbone=Qwen2.52024.10 | 92.77 | 83.37 | |
| LoRAAttack=InSent, Backbone=Qwen2.52024.10 | 92.65 | 100 | |
| ONIONAttack=InSent, Backbone=LLaMA32024.10 | 92.64 | 98.33 | |
| PruneAttack=BadNet, Backbone=Qwen2.52024.10 | 92.39 | 7.9 | |
| LoRAAttack=InSent, Backbone=Vicuna2024.10 | 92.39 | 82.33 | |
| W2SDefenseAttack=BadNet, Backbone=Qwen2.52024.10 | 92.13 | 2.91 | |
| Back Tr.Attack=BadNet, Backbone=Qwen2.52024.10 | 92 | 36.17 | |
| W2SDefenseAttack=SynAttack, Backbone=Vicuna2024.10 | 91.87 | 8.96 | |
| ONIONAttack=SynAttack, Backbone=Qwen2.52024.10 | 91.87 | 23.54 | |
| LoRAAttack=SynAttack, Backbone=Vicuna2024.10 | 91.74 | 17.29 | |
| W2SDefenseAttack=InSent, Backbone=Vicuna2024.10 | 91.61 | 10.6 | |
| W2SDefenseAttack=InSent, Backbone=Qwen2.52024.10 | 91.61 | 4.99 | |
| W2SDefenseAttack=InSent, Backbone=LLaMA32024.10 | 91.48 | 17.88 | |
| PruneAttack=SynAttack, Backbone=Vicuna2024.10 | 91.48 | 22.71 | |
| Back Tr.Attack=SynAttack, Backbone=Qwen2.52024.10 | 91.48 | 22.5 | |
| Back Tr.Attack=BadNet, Backbone=Vicuna2024.10 | 91.35 | 42.2 | |
| PruneAttack=BadNet, Backbone=Vicuna2024.10 | 91.23 | 35.14 | |
| W2SDefenseAttack=SynAttack, Backbone=LLaMA32024.10 | 90.97 | 15.83 | |
| PruneAttack=InSent, Backbone=Vicuna2024.10 | 90.71 | 50.73 | |
| ONIONAttack=BadNet, Backbone=LLaMA32024.10 | 90.45 | 30.56 | |
| ONIONAttack=BadNet, Backbone=Qwen2.52024.10 | 90.45 | 26.4 | |
| Back Tr.Attack=InSent, Backbone=Vicuna2024.10 | 90.32 | 81.7 | |
| ONIONAttack=InSent, Backbone=Qwen2.52024.10 | 90.19 | 98.75 | |
| W2SDefenseAttack=SynAttack, Backbone=Qwen2.52024.10 | 90.06 | 15.83 | |
| ONIONAttack=InSent, Backbone=Vicuna2024.10 | 89.93 | 88.77 | |
| ONIONAttack=SynAttack, Backbone=Vicuna2024.10 | 89.8 | 18.33 | |
| ONIONAttack=BadNet, Backbone=Vicuna2024.10 | 88.9 | 32.64 | |
| Back Tr.Attack=SynAttack, Backbone=LLaMA32024.10 | 86.58 | 18.96 | |
| LoRAAttack=SynAttack, Backbone=LLaMA32024.10 | 86.45 | 21.25 | |
| PruneAttack=SynAttack, Backbone=LLaMA32024.10 | 85.68 | 21.88 | |
| ONIONAttack=SynAttack, Backbone=LLaMA32024.10 | 83.61 | 26.66 | |
| SCPDAttack=BadNet, Backbone=Qwen2.52024.10 | 83.42 | 34.58 | |
| SCPDAttack=InSent, Backbone=Qwen2.52024.10 | 83.42 | 21.46 | |
| SCPDAttack=InSent, Backbone=LLaMA32024.10 | 82.51 | 32.29 | |
| SCPDAttack=SynAttack, Backbone=Qwen2.52024.10 | 82.51 | 17.08 | |
| SCPDAttack=InSent, Backbone=Vicuna2024.10 | 82.25 | 18.54 | |
| SCPDAttack=BadNet, Backbone=LLaMA32024.10 | 81.61 | 35.21 | |
| SCPDAttack=SynAttack, Backbone=Vicuna2024.10 | 81.48 | 12.71 | |
| SCPDAttack=BadNet, Backbone=Vicuna2024.10 | 81.35 | 40 | |
| PruneAttack=SynAttack, Backbone=Qwen2.52024.10 | 80.39 | 33.13 | |
| SCPDAttack=SynAttack, Backbone=LLaMA32024.10 | 79.02 | 20 | |
| PruneAttack=InSent, Backbone=Qwen2.52024.10 | 76 | 24.53 | |
| Back Tr.Attack=SynAttack, Backbone=Vicuna2024.10 | 66.45 | 81.46 |