Sentiment Classification on SST-2 64 instances (test)
92.55AccuracyBefore
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BeforeBackdoor Attack=Word+EP2022.10 | 92.55 | 100 | |
| BeforeBackdoor Attack=Sentence (Scratch)2022.10 | 92.32 | 100 | |
| BeforeBackdoor Attack=Trigger Sentence2022.10 | 92.2 | 100 | |
| Fine-pruningBackdoor Attack=Word+EP2022.10 | 92.2 | 100 | |
| BeforeBackdoor Attack=Word (Scratch)2022.10 | 92.09 | 100 | |
| BeforeBackdoor Attack=Logit Anchoring2022.10 | 92.09 | 100 | |
| STRIPBackdoor Attack=Logit Anchoring2022.10 | 92.09 | 21.4 | |
| BeforeBackdoor Attack=Logit Anchoring2022.10 | 92.09 | 100 | |
| Fine-tuningBackdoor Attack=Logit Anchoring2022.10 | 92.08 | 100 | |
| Fine-tuningBackdoor Attack=Adaptive Attack2022.10 | 91.97 | 100 | |
| BeforeBackdoor Attack=Layer-wise Attack2022.10 | 91.86 | 100 | |
| Fine-pruningBackdoor Attack=Word (Scratch)2022.10 | 91.86 | 100 | |
| BeforeBackdoor Attack=Layer-wise Attack2022.10 | 91.86 | 100 | |
| Fine-pruningBackdoor Attack=Trigger Sentence2022.10 | 91.63 | 100 | |
| BeforeBackdoor Attack=Syntactic Trigger2022.10 | 91.52 | 97.52 | |
| Fine-pruningBackdoor Attack=Sentence (Scratch)2022.10 | 91.4 | 100 | |
| BeforeBackdoor Attack=Word+ES (Scratch)2022.10 | 91.28 | 100 | |
| ONIONBackdoor Attack=Word (Scratch)2022.10 | 91.28 | 54.5 | |
| ONIONBackdoor Attack=Trigger Sentence2022.10 | 91.28 | 98.87 | |
| BeforeBackdoor Attack=Adaptive Attack2022.10 | 91.28 | 100 | |
| STRIPBackdoor Attack=Trigger Sentence2022.10 | 91.17 | 19.37 | |
| Fine-tuningBackdoor Attack=Layer-wise Attack2022.10 | 91.06 | 100 | |
| Fine-mixing (Sel)Backdoor Attack=Layer-wise Attack2022.10 | 91.05 | 42.79 | |
| ONIONBackdoor Attack=Word+ES (Scratch)2022.10 | 90.9 | 54.73 | |
| Fine-pruningBackdoor Attack=Word+ES2022.10 | 90.83 | 100 | |
| Fine-pruningBackdoor Attack=Syntactic Trigger2022.10 | 90.71 | 96.62 | |
| STRIPBackdoor Attack=Syntactic Trigger2022.10 | 90.71 | 97.52 | |
| Fine-mixing (Sel)Backdoor Attack=Adaptive Attack2022.10 | 90.6 | 59.46 | |
| RAPBackdoor Attack=Word+EP2022.10 | 90.37 | 100 | |
| Fine-pruningBackdoor Attack=Adaptive Attack2022.10 | 90.37 | 100 | |
| STRIPBackdoor Attack=Word+EP2022.10 | 90.25 | 100 | |
| RAPBackdoor Attack=Word (Scratch)2022.10 | 90.25 | 20.27 | |
| BeforeBackdoor Attack=Word+ES2022.10 | 90.14 | 100 | |
| STRIPBackdoor Attack=Layer-wise Attack2022.10 | 90.14 | 28.6 | |
| Fine-pruningBackdoor Attack=Trigger Word2022.10 | 90.02 | 100 | |
| Fine-pruningBackdoor Attack=Word+ES (Scratch)2022.10 | 90.02 | 100 | |
| RAPBackdoor Attack=Sentence (Scratch)2022.10 | 90.02 | 40.54 | |
| Fine-pruningBackdoor Attack=Trigger Word2022.10 | 90.02 | 100 | |
| Fine-mixingBackdoor Attack=Adaptive Attack2022.10 | 90.02 | 21.85 | |
| BeforeBackdoor Attack=Trigger Word2022.10 | 89.79 | 100 | |
| Fine-mixingBackdoor Attack=Word+ES (Scratch)2022.10 | 89.79 | 13.06 | |
| Fine-mixingBackdoor Attack=Layer-wise Attack2022.10 | 89.79 | 15.77 | |
| Fine-mixingBackdoor Attack=Logit Anchoring2022.10 | 89.79 | 16.22 | |
| BeforeBackdoor Attack=Trigger Word2022.10 | 89.79 | 100 | |
| Fine-mixingBackdoor Attack=Layer-wise Attack2022.10 | 89.79 | 15.77 | |
| Fine-mixingBackdoor Attack=Logit Anchoring2022.10 | 89.79 | 16.22 | |
| ONIONBackdoor Attack=Word+EP2022.10 | 89.68 | 20.32 | |
| ONIONBackdoor Attack=Sentence (Scratch)2022.10 | 89.68 | 71.4 | |
| STRIPBackdoor Attack=Word+ES (Scratch)2022.10 | 89.68 | 25.9 | |
| ONIONBackdoor Attack=Word+ES2022.10 | 89.56 | 53.38 | |
| RAPBackdoor Attack=Syntactic Trigger2022.10 | 89.56 | 94.37 | |
| RAPBackdoor Attack=Logit Anchoring2022.10 | 89.56 | 17.79 | |
| Fine-mixingBackdoor Attack=Word+EP2022.10 | 89.56 | 14.25 | |
| Fine-mixingBackdoor Attack=Trigger Word2022.10 | 89.45 | 14.19 | |
| Fine-mixingBackdoor Attack=Word (Scratch)2022.10 | 89.45 | 22.75 | |
| Fine-mixingBackdoor Attack=Sentence (Scratch)2022.10 | 89.45 | 17.79 | |
| Fine-mixingBackdoor Attack=Trigger Word2022.10 | 89.45 | 14.19 | |
| Fine-mixingBackdoor Attack=Trigger Sentence2022.10 | 89.44 | 17.78 | |
| Fine-pruningBackdoor Attack=Layer-wise Attack2022.10 | 89.33 | 100 | |
| ONIONBackdoor Attack=Layer-wise Attack2022.10 | 89.33 | 11.04 | |
| STRIPBackdoor Attack=Word (Scratch)2022.10 | 89.33 | 22.3 | |
| RAPBackdoor Attack=Word+ES (Scratch)2022.10 | 89.33 | 21.62 | |
| Fine-tuningBackdoor Attack=Trigger Word2022.10 | 89.33 | 100 | |
| Fine-pruningBackdoor Attack=Layer-wise Attack2022.10 | 89.33 | 100 | |
| Fine-pruningBackdoor Attack=Logit Anchoring2022.10 | 89.22 | 100 | |
| RAPBackdoor Attack=Trigger Sentence2022.10 | 89.22 | 24.55 | |
| Fine-mixingBackdoor Attack=Word+ES2022.10 | 89.22 | 14.64 | |
| Fine-mixingBackdoor Attack=Syntactic Trigger2022.10 | 89.22 | 22.07 | |
| Fine-mixing (Sel)Backdoor Attack=Trigger Word2022.10 | 89.22 | 15.77 | |
| Fine-pruningBackdoor Attack=Logit Anchoring2022.10 | 89.22 | 100 | |
| Fine-mixing (Sel)Backdoor Attack=Logit Anchoring2022.10 | 89.22 | 28.38 | |
| ONIONBackdoor Attack=Logit Anchoring2022.10 | 89.11 | 11.03 | |
| STRIPBackdoor Attack=Sentence (Scratch)2022.10 | 89.11 | 16.67 | |
| RAPBackdoor Attack=Layer-wise Attack2022.10 | 89.11 | 18.7 | |
| ONIONBackdoor Attack=Syntactic Trigger2022.10 | 89.1 | 93.02 | |
| ONIONBackdoor Attack=Trigger Word2022.10 | 88.53 | 54.73 | |
| RAPBackdoor Attack=Trigger Word2022.10 | 84.52 | 14.86 | |
| RAPBackdoor Attack=Word+ES2022.10 | 81.54 | 10.59 | |
| STRIPBackdoor Attack=Trigger Word2022.10 | 75.11 | 11.04 | |
| STRIPBackdoor Attack=Word+ES2022.10 | 71.22 | 8.38 |