Backdoor Trigger Detection on SST-2
10RecallCSO(continuous)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=high2026.06 | 10 | — | — | 1 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=high2026.06 | 10 | — | — | 1 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=high2026.06 | 10 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=high2026.06 | 10 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=high2026.06 | 10 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=low2026.06 | 10 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=high2026.06 | 10 | — | — | 1 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=low2026.06 | 10 | — | — | 0 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=high2026.06 | 10 | — | — | 0 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=high2026.06 | 10 | — | — | 0 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=high2026.06 | 9 | — | — | 1 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=low2026.06 | 9 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=low2026.06 | 9 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=low2026.06 | 9 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=high2026.06 | 9 | — | — | 1 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=high2026.06 | 9 | — | — | 0 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=low2026.06 | 9 | — | — | 0 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=low2026.06 | 8 | — | — | 1 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=low2026.06 | 8 | — | — | 1 | |
| CSO(discrete)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=low2026.06 | 8 | — | — | 1 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=high2026.06 | 8 | — | — | 0 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=low2026.06 | 8 | — | — | 0 | |
| CSO(continuous)Model=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=low2026.06 | 7 | — | — | 1 | |
| DBSModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=high2026.06 | 7 | — | — | 10 | |
| PICCOLOModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=low2026.06 | 7 | — | — | 0 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=high2026.06 | 7 | — | — | 0 | |
| DBSModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=low2026.06 | 5 | — | — | 10 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=low2026.06 | 5 | — | — | 0 | |
| DBSModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=low2026.06 | 4 | — | — | 10 | |
| DBSModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=high2026.06 | 3 | — | — | 10 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=high2026.06 | 3 | — | — | 0 | |
| DBSModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=high2026.06 | 2 | — | — | 10 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=low2026.06 | 2 | — | — | 0 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=low2026.06 | 2 | — | — | 0 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=high2026.06 | 2 | — | — | 0 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=low2026.06 | 1 | — | — | 0 | |
| MMModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=low2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=high2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=low2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=dirty label, Poison Rate=high2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=low2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=high2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=low2026.06 | 0 | — | — | 8 | |
| MMModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=dirty label, Poison Rate=high2026.06 | 0 | — | — | 8 | |
| DBSModel=qwen3-0.6b, Trigger Type=Tell me seriously, Poison Label Type=clean label, Poison Rate=low2026.06 | 0 | — | — | 10 | |
| DBSModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=low2026.06 | 0 | — | — | 10 | |
| DBSModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=high2026.06 | 0 | — | — | 10 | |
| CLIBEModel=qwen3-0.6b, Trigger Type=mb, Poison Label Type=clean label, Poison Rate=high2026.06 | 0 | — | — | 0 | |
| DANAttack=badnets2024.05 | — | — | 70.42 | — | |
| DANAttack=addsent2024.05 | — | — | 64.63 | — | |
| DANAttack=stylebkd2024.05 | — | — | 72.94 | — | |
| DANAttack=synbkd2024.05 | — | — | 79.11 | — | |
| NASAttack=badnets2024.05 | — | — | 98.77 | — | |
| NASAttack=addsent2024.05 | — | — | 97.96 | — | |
| NASAttack=stylebkd2024.05 | — | — | 87.37 | — | |
| NASAttack=synbkd2024.05 | — | — | 88.83 | — | |
| RAPAttack=badnets2024.05 | — | — | 64.22 | — | |
| RAPAttack=addsent2024.05 | — | — | 70.57 | — | |
| RAPAttack=stylebkd2024.05 | — | — | 52.42 | — | |
| RAPAttack=synbkd2024.05 | — | — | 59.89 | — | |
| STRIPAttack=badnets2024.05 | — | — | 52.63 | — | |
| STRIPAttack=addsent2024.05 | — | — | 51.68 | — | |
| STRIPAttack=stylebkd2024.05 | — | — | 53.78 | — | |
| STRIPAttack=synbkd2024.05 | — | — | 50.51 | — |