Sentiment Analysis on SST2 (CACC, ASR)
99ASRUnpatched
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UnpatchedAttack=Sleeper2026.06 | 99 | — | 95 | |
| UnpatchedAttack=POLISHED2026.06 | 97 | — | 95 | |
| FTℓ2Attack=POLISHED2026.06 | 97 | — | 94 | |
| FinePruningAttack=POLISHED2026.06 | 97 | — | 94 | |
| BAERASERAttack=POLISHED2026.06 | 97 | — | 90 | |
| MudjackingAttack=POLISHED2026.06 | 97 | — | 90 | |
| SPPAttack=POLISHED2026.06 | 97 | — | 91 | |
| OneShotAttack=POLISHED2026.06 | 97 | — | 90 | |
| UnpatchedAttack=CBA2026.06 | 97 | — | 95 | |
| FTAttack=CBA2026.06 | 97 | — | 90 | |
| FT_l2Attack=CBA2026.06 | 97 | — | 91 | |
| FT_linfAttack=CBA2026.06 | 97 | — | 91 | |
| FinePruningAttack=CBA2026.06 | 97 | — | 94 | |
| OneShotAttack=CBA2026.06 | 97 | — | 94 | |
| MENDAttack=CBA2026.06 | 97 | — | 93 | |
| UnpatchedAttack=POLISHED2026.06 | 97 | — | 94 | |
| UnpatchedAttack=CBA2026.06 | 96 | — | 95 | |
| FTAttack=CBA2026.06 | 96 | — | 94 | |
| FTAttack=POLISHED2026.06 | 96 | — | 94 | |
| FTℓ2Attack=CBA2026.06 | 96 | — | 91 | |
| FTℓ∞Attack=CBA2026.06 | 96 | — | 92 | |
| FTℓ∞Attack=POLISHED2026.06 | 96 | — | 94 | |
| FinePruningAttack=CBA2026.06 | 96 | — | 94 | |
| BAERASERAttack=CBA2026.06 | 96 | — | 91 | |
| SPPAttack=CBA2026.06 | 96 | — | 92 | |
| OneShotAttack=CBA2026.06 | 96 | — | 94 | |
| MENDAttack=CBA2026.06 | 96 | — | 93 | |
| MENDAttack=POLISHED2026.06 | 96 | — | 93 | |
| UnpatchedAttack=UJB2026.06 | 96 | — | 95 | |
| FT_l2Attack=UJB2026.06 | 96 | — | 93 | |
| FinePruningAttack=UJB2026.06 | 96 | — | 92 | |
| SPPAttack=UJB2026.06 | 96 | — | 90 | |
| OneShotAttack=UJB2026.06 | 96 | — | 93 | |
| MENDAttack=UJB2026.06 | 96 | — | 93 | |
| UnpatchedAttack=UJB2026.06 | 95 | — | 95 | |
| FTAttack=UJB2026.06 | 95 | — | 90 | |
| FTℓ2Attack=UJB2026.06 | 95 | — | 90 | |
| FTℓ∞Attack=UJB2026.06 | 95 | — | 94 | |
| FinePruningAttack=UJB2026.06 | 95 | — | 94 | |
| BAERASERAttack=UJB2026.06 | 95 | — | 90 | |
| MENDAttack=UJB2026.06 | 95 | — | 93 | |
| ROMEAttack=POLISHED2026.06 | 95 | — | 92 | |
| UnpatchedAttack=BadNets2026.06 | 95 | — | 95 | |
| FT_l2Attack=BadNets2026.06 | 95 | — | 93 | |
| MENDAttack=BadNets2026.06 | 95 | — | 94 | |
| BAERASERAttack=CBA2026.06 | 95 | — | 91 | |
| ROMEAttack=CBA2026.06 | 95 | — | 94 | |
| UnpatchedAttack=BadNets2026.06 | 94 | — | 95 | |
| FTAttack=BadNets2026.06 | 94 | — | 93 | |
| FTℓ2Attack=BadNets2026.06 | 94 | — | 91 | |
| FTℓ∞Attack=BadNets2026.06 | 94 | — | 91 | |
| FinePruningAttack=BadNets2026.06 | 94 | — | 91 | |
| BAERASERAttack=BadNets2026.06 | 94 | — | 91 | |
| SPPAttack=UJB2026.06 | 94 | — | 94 | |
| SPPAttack=BadNets2026.06 | 94 | — | 90 | |
| OneShotAttack=UJB2026.06 | 94 | — | 92 | |
| OneShotAttack=BadNets2026.06 | 94 | — | 90 | |
| MENDAttack=BadNets2026.06 | 94 | — | 93 | |
| ROMEAttack=CBA2026.06 | 94 | — | 94 | |
| FinePruningAttack=BadNets2026.06 | 94 | — | 94 | |
| OneShotAttack=BadNets2026.06 | 94 | — | 92 | |
| ROMEAttack=BadNets2026.06 | 94 | — | 93 | |
| SPPAttack=CBA2026.06 | 94 | — | 94 | |
| MudjackingAttack=UJB2026.06 | 93 | — | 94 | |
| ROMEAttack=BadNets2026.06 | 93 | — | 90 | |
| FTAttack=UJB2026.06 | 93 | — | 93 | |
| BAERASERAttack=UJB2026.06 | 93 | — | 91 | |
| MudjackingAttack=BadNets2026.06 | 92 | — | 93 | |
| FT_linfAttack=UJB2026.06 | 92 | — | 94 | |
| ROMEAttack=UJB2026.06 | 92 | — | 93 | |
| FTAttack=BadNets2026.06 | 92 | — | 94 | |
| FT_linfAttack=BadNets2026.06 | 91 | — | 94 | |
| SPPAttack=BadNets2026.06 | 90 | — | 94 | |
| UnpatchedAttack=Sleeper2026.06 | 89 | — | 95 | |
| FTAttack=Sleeper2026.06 | 89 | — | 93 | |
| FTℓ2Attack=Sleeper2026.06 | 89 | — | 93 | |
| FTℓ∞Attack=Sleeper2026.06 | 89 | — | 94 | |
| ROMEAttack=UJB2026.06 | 89 | — | 90 | |
| MENDAttack=Sleeper2026.06 | 88 | — | 94 | |
| FinePruningAttack=Sleeper2026.06 | 87 | — | 90 | |
| ROMEAttack=Sleeper2026.06 | 87 | — | 92 | |
| BAERASERAttack=BadNets2026.06 | 86 | — | 94 | |
| SPPAttack=Sleeper2026.06 | 85 | — | 93 | |
| BAERASERAttack=Sleeper2026.06 | 82 | — | 91 | |
| OneShotAttack=Sleeper2026.06 | 78 | — | 92 | |
| MudjackingAttack=CBA2026.06 | 66 | — | 93 | |
| MudjackingAttack=UJB2026.06 | 34 | — | 92 | |
| MudjackingAttack=CBA2026.06 | 32 | — | 90 | |
| MudjackingAttack=Sleeper2026.06 | 30 | — | 91 | |
| Test-time DefenseTrigger=Phrase, Model=Llama3-8B-Instruct2025.10 | 23.7 | 95.8 | — | |
| SPPAttack=Sleeper2026.06 | 20 | — | 94 | |
| Test-time DefenseTrigger=Long, Model=Llama3-8B-Instruct2025.10 | 17.85 | 94.32 | — | |
| Test-time DefenseTrigger=Word, Model=Llama3-8B-Instruct2025.10 | 10.25 | 95.22 | — | |
| MudjackingAttack=BadNets2026.06 | 10 | — | 92 | |
| LocphylaxTrigger=Word, Model=Llama3-8B-Instruct2025.10 | 7.5 | 96.2 | — | |
| SPPAttack=No attack2026.06 | 6 | — | 94 | |
| FTAttack=No attack2026.06 | 5 | — | 94 | |
| FinePruningAttack=No attack2026.06 | 5 | — | 94 | |
| MENDAttack=No attack2026.06 | 4 | — | 93 | |
| UnpatchedAttack=No attack2026.06 | 3 | — | 95 |