List your company

Misclassification Detection on Cola

81.6ROC-AUCALIEN
1.83222.54143.2563.959May 21, 2025

Evaluation results

MethodLinks
ALIENBackbone=Qwen2.5 7B, Detection Method=ALIEN2025.0581.6
ALIENBackbone=Roberta, Detection Method=ALIEN2025.0580.5
ALIENBackbone=Electra, Detection Method=ALIEN2025.0578.8
ALIENBackbone=LLaMA2 7B, Detection Method=ALIEN2025.0578
Probability based (Entropy)Backbone=LLaMA2 7B, Detection Method=Probability based, Method Variant=Entropy2025.0577
Probability based (SR)Backbone=LLaMA2 7B, Detection Method=Probability based, Method Variant=SR2025.0576.9
ALIENBackbone=Qwen3 1.7B, Detection Method=ALIEN2025.0573.7
Linear probing (Last)Backbone=LLaMA2 7B, Detection Method=Linear probing, Layer Selection=Last2025.0572
Mahalanobis distance (MDR)Backbone=LLaMA2 7B, Detection Method=Mahalanobis distance, Method Variant=MDR2025.0569.9
RDEBackbone=LLaMA2 7B, Detection Method=RDE2025.0568
Mahalanobis distance (MD)Backbone=LLaMA2 7B, Detection Method=Mahalanobis distance, Method Variant=MD2025.0562.2
Mahalanobis distance (MDM)Backbone=LLaMA2 7B, Detection Method=Mahalanobis distance, Method Variant=MDM2025.0562.2
Attention pooling (Last)Backbone=LLaMA2 7B, Detection Method=Attention pooling, Layer Selection=Last2025.0561.5
Linear probing (Mid)Backbone=LLaMA2 7B, Detection Method=Linear probing, Layer Selection=Mid2025.0560.6
Attention pooling (Begin)Backbone=LLaMA2 7B, Detection Method=Attention pooling, Layer Selection=Begin2025.0560.3
Linear probing (Begin)Backbone=LLaMA2 7B, Detection Method=Linear probing, Layer Selection=Begin2025.0558.1
Attention pooling (Mid)Backbone=LLaMA2 7B, Detection Method=Attention pooling, Layer Selection=Mid2025.0551.4
MDBackbone=Qwen2.5 7B2025.0514.5
Attention PoolingBackbone=LLaMA2 7B, Layer=Last2025.0511.2
MDMBackbone=LLaMA2 7B2025.0510.4
MDBackbone=LLaMA2 7B2025.0510.3
MDRBackbone=LLaMA2 7B2025.059.3
ALIENBackbone=Qwen3 1.7B2025.059
RDEBackbone=LLaMA2 7B2025.058.4
Linear ProbingBackbone=LLaMA2 7B, Layer=Last2025.057.5
SRBackbone=LLaMA2 7B2025.056.8
EntropyBackbone=LLaMA2 7B2025.056.5
ALIENBackbone=LLaMA2 7B2025.056.2
ALIENBackbone=Roberta2025.056
ALIENBackbone=Qwen2.5 7B2025.055.3
ALIENBackbone=Electra2025.054.9