Backdoor Attack on Fear Trigger Emotion (Generalization)
79.8ASR (Generalization)GREAT
Evaluation Results
| Method | Links | |
|---|---|---|
| GREATModel=Llama-3.2-3B, Poisoning ratio (α)=10%, K=20002025.10 | 79.8 | |
| GREATModel=Llama-3.2-1B, Poisoning ratio (α)=10%, K=20002025.10 | 77.2 | |
| GREATModel=Llama-3.2-3B, Poisoning ratio (α)=1%, K=1002025.10 | 71 | |
| GREATModel=Llama-3.2-3B, Poisoning ratio (α)=10%, K=12025.10 | 62 | |
| GREATModel=Llama-3.2-1B, Poisoning ratio (α)=10%, K=12025.10 | 61 | |
| RandomModel=Llama-3.2-1B, Poisoning ratio (α)=10%, K=20002025.10 | 60.2 | |
| RandomModel=Llama-3.2-3B, Poisoning ratio (α)=10%, K=20002025.10 | 60.1 | |
| GREATModel=Llama-3.2-1B, Poisoning ratio (α)=1%, K=1002025.10 | 56.2 | |
| RandomModel=Llama-3.2-1B, Poisoning ratio (α)=10%, K=12025.10 | 55.5 | |
| RandomModel=Llama-3.2-3B, Poisoning ratio (α)=10%, K=12025.10 | 55.3 | |
| RandomModel=Llama-3.2-3B, Poisoning ratio (α)=1%, K=1002025.10 | 55 | |
| GREATModel=Llama-3.2-3B, Poisoning ratio (α)=1%, K=12025.10 | 53.8 | |
| SUDOModel=Llama-3.2-1B, Poisoning ratio (α)=10%2025.10 | 50.5 | |
| GREATModel=Llama-3.2-1B, Poisoning ratio (α)=1%, K=12025.10 | 50.2 | |
| SUDOModel=Llama-3.2-3B, Poisoning ratio (α)=10%2025.10 | 50 | |
| RandomModel=Llama-3.2-1B, Poisoning ratio (α)=1%, K=1002025.10 | 48 | |
| RandomModel=Llama-3.2-3B, Poisoning ratio (α)=1%, K=12025.10 | 45.8 | |
| SUDOModel=Llama-3.2-3B, Poisoning ratio (α)=1%2025.10 | 44.8 | |
| RandomModel=Llama-3.2-1B, Poisoning ratio (α)=1%, K=12025.10 | 44 | |
| SUDOModel=Llama-3.2-1B, Poisoning ratio (α)=1%2025.10 | 43.2 |