Sentiment Classification on SST-2 (Repair and Attack Metrics)
99.67Attack Generation RateVanilla
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VanillaVictim=BERT, Attack=BadNets2025.11 | 99.67 | — | — | — | 92.65 | |
| VanillaVictim=BERT, Attack=Cbat2025.11 | 95.42 | — | — | — | 90.85 | |
| VanillaVictim=BERT, Attack=HiddenKiller2025.11 | 94.16 | — | — | — | 89.33 | |
| VanillaVictim=BERT, Attack=BGMAttack2025.11 | 91.27 | — | — | — | 88.21 | |
| VanillaVictim=BERT, Attack=NWS2025.11 | 90.23 | — | — | — | 90.45 | |
| VanillaVictim=Llama2-7B, Attack=NWS2025.11 | 89.25 | — | — | — | 85.15 | |
| VanillaVictim=Llama2-7B, Attack=BadNets2025.11 | 88.73 | — | — | — | 86.91 | |
| VanillaVictim=Llama2-7B, Attack=Cbat2025.11 | 85.31 | — | — | — | 84.01 | |
| VanillaVictim=Llama2-7B, Attack=BGMAttack2025.11 | 84.19 | — | — | — | 83.2 | |
| VanillaVictim=Llama2-7B, Attack=HiddenKiller2025.11 | 83.27 | — | — | — | 81.56 | |
| WARPModel=DistilBERT2026.04 | 69 | 100 | 100 | 91.4 | — | |
| WARPModel=BERT2026.04 | 64.7 | 100 | 100 | 92.7 | — | |
| Same-layer LoRAModel=BERT, rank=22026.04 | 60 | 91.9 | 99.6 | 98 | — | |
| Same-layer LoRAModel=DistilBERT, rank=22026.04 | 56.7 | 75.9 | 99 | 97.5 | — | |
| All-layers Full FTModel=BERT2026.04 | 56.7 | 100 | 99.5 | 93.8 | — | |
| All-layers Full FTModel=DistilBERT2026.04 | 50.2 | 100 | 99.5 | 93.6 | — | |
| FPVictim=Llama2-7B, Attack=Cbat2025.11 | 50.14 | — | — | — | 80.97 | |
| FPVictim=Llama2-7B, Attack=HiddenKiller2025.11 | 48.27 | — | — | — | 80.1 | |
| FPVictim=Llama2-7B, Attack=BadNets2025.11 | 46.39 | — | — | — | 83.19 | |
| PUREVictim=Llama2-7B, Attack=HiddenKiller2025.11 | 40.24 | — | — | — | 79.39 | |
| PUREVictim=Llama2-7B, Attack=Cbat2025.11 | 39.49 | — | — | — | 80.35 | |
| FPVictim=Llama2-7B, Attack=BGMAttack2025.11 | 39.33 | — | — | — | 82.78 | |
| FPVictim=BERT, Attack=Cbat2025.11 | 39.23 | — | — | — | 89.93 | |
| MEFTVictim=BERT, Attack=HiddenKiller2025.11 | 38.05 | — | — | — | 89.21 | |
| FPVictim=BERT, Attack=BGMAttack2025.11 | 37.15 | — | — | — | 86.17 | |
| FPVictim=BERT, Attack=HiddenKiller2025.11 | 36.45 | — | — | — | 89.17 | |
| MEFTVictim=BERT, Attack=Cbat2025.11 | 36.12 | — | — | — | 91.06 | |
| PUREVictim=Llama2-7B, Attack=BGMAttack2025.11 | 35.45 | — | — | — | 80.28 | |
| PUREVictim=BERT, Attack=HiddenKiller2025.11 | 35.26 | — | — | — | 88.39 | |
| PUREVictim=Llama2-7B, Attack=BadNets2025.11 | 33.45 | — | — | — | 82.37 | |
| FPVictim=Llama2-7B, Attack=NWS2025.11 | 32.04 | — | — | — | 81.45 | |
| PUREVictim=BERT, Attack=BGMAttack2025.11 | 29.46 | — | — | — | 81.27 | |
| PUREVictim=Llama2-7B, Attack=NWS2025.11 | 28.44 | — | — | — | 83.61 | |
| PUREVictim=BERT, Attack=Cbat2025.11 | 28.23 | — | — | — | 89.76 | |
| MEFTVictim=Llama2-7B, Attack=HiddenKiller2025.11 | 28.17 | — | — | — | 80.78 | |
| FPVictim=BERT, Attack=BadNets2025.11 | 26.36 | — | — | — | 92.28 | |
| MEFTVictim=Llama2-7B, Attack=BadNets2025.11 | 26.34 | — | — | — | 85.17 | |
| Attention Safety AlignmentVictim=Llama2-7B, Attack=BadNets2025.11 | 25.61 | — | — | — | 85.19 | |
| FPVictim=BERT, Attack=NWS2025.11 | 25.33 | — | — | — | 87.19 | |
| MEFTVictim=Llama2-7B, Attack=BGMAttack2025.11 | 24.03 | — | — | — | 81.3 | |
| MEFTVictim=Llama2-7B, Attack=NWS2025.11 | 23.91 | — | — | — | 82.19 | |
| OriginalModel=BERT2026.04 | 23.2 | 1 | 98.2 | 98.5 | — | |
| Attention Safety AlignmentVictim=Llama2-7B, Attack=Cbat2025.11 | 23.14 | — | — | — | 83.19 | |
| Attention Safety AlignmentVictim=Llama2-7B, Attack=BGMAttack2025.11 | 23.07 | — | — | — | 82.95 | |
| MEFTVictim=BERT, Attack=BGMAttack2025.11 | 22.75 | — | — | — | 84.31 | |
| MEFTVictim=Llama2-7B, Attack=Cbat2025.11 | 22.55 | — | — | — | 80.13 | |
| OriginalModel=DistilBERT2026.04 | 22.3 | 5.6 | 98.1 | 98.2 | — | |
| Attention Safety AlignmentVictim=BERT, Attack=BGMAttack2025.11 | 21.73 | — | — | — | 85.58 | |
| Attention Safety AlignmentVictim=BERT, Attack=Cbat2025.11 | 20.36 | — | — | — | 91.45 | |
| Attention Safety AlignmentVictim=Llama2-7B, Attack=HiddenKiller2025.11 | 20.16 | — | — | — | 82.04 | |
| PUREVictim=BERT, Attack=NWS2025.11 | 20.15 | — | — | — | 88.3 | |
| MEFTVictim=BERT, Attack=NWS2025.11 | 19.67 | — | — | — | 89.31 | |
| Attention Safety AlignmentVictim=Llama2-7B, Attack=NWS2025.11 | 18.33 | — | — | — | 83.16 | |
| PUREVictim=BERT, Attack=BadNets2025.11 | 16.98 | — | — | — | 90.64 | |
| Attention Safety AlignmentVictim=BERT, Attack=HiddenKiller2025.11 | 15.66 | — | — | — | 90.03 | |
| Attention Safety AlignmentVictim=BERT, Attack=NWS2025.11 | 15.48 | — | — | — | 90.54 | |
| MEFTVictim=BERT, Attack=BadNets2025.11 | 15.23 | — | — | — | 91.48 | |
| Attention Safety AlignmentVictim=BERT, Attack=BadNets2025.11 | 11.23 | — | — | — | 91.09 |