Backdoor mitigation on Lingspam
99.65Clean AccuracyVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaAttack=Badnets, Target LM=BERT-base2024.02 | 99.65 | 3.31 | |
| VanillaAttack=Addsent, Target LM=BERT-base2024.02 | 99.65 | 86.11 | |
| VanillaAttack=HiddenKiller, Target LM=BERT-base2024.02 | 99.31 | 98.97 | |
| VanillaAttack=StyleBkd, Target LM=BERT-base2024.02 | 98.96 | 92.24 | |
| PrefixAttack=HiddenKiller, Target LM=BERT-base2024.02 | 96.21 | 86.92 | |
| PrefixAttack=Badnets, Target LM=BERT-base2024.02 | 96.03 | 2.27 | |
| MuScleLoRAAttack=HiddenKiller, Target LM=BERT-base2024.02 | 95.52 | 0.2 | |
| LoRAAttack=StyleBkd, Target LM=BERT-base2024.02 | 95.17 | 37.1 | |
| MuScleLoRAAttack=StyleBkd, Target LM=BERT-base2024.02 | 93.96 | 4.28 | |
| PrefixAttack=StyleBkd, Target LM=BERT-base2024.02 | 93.79 | 8.59 | |
| LoRAAttack=HiddenKiller, Target LM=BERT-base2024.02 | 92.93 | 27.69 | |
| MuScleLoRAAttack=Badnets, Target LM=BERT-base2024.02 | 91.89 | 0 | |
| MuScleLoRAAttack=Addsent, Target LM=BERT-base2024.02 | 90.68 | 1.24 | |
| PrefixAttack=Addsent, Target LM=BERT-base2024.02 | 90.51 | 4.35 | |
| LoRAAttack=Addsent, Target LM=BERT-base2024.02 | 90.34 | 1.24 | |
| AdapterAttack=Badnets, Target LM=BERT-base2024.02 | 86.55 | 2.69 | |
| AdapterAttack=Addsent, Target LM=BERT-base2024.02 | 85.69 | 7.45 | |
| LoRAAttack=Badnets, Target LM=BERT-base2024.02 | 85.17 | 0 | |
| AdapterAttack=StyleBkd, Target LM=BERT-base2024.02 | 84.66 | 2.1 | |
| AdapterAttack=HiddenKiller, Target LM=BERT-base2024.02 | 83.79 | 1.05 |