OOD Detection on WMT15 En-Fr (ID) IT (OOD) (test)
97.41AUROCAP-OOD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AP-OODOOD level=Input, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 97.41 | 16.89 | |
| Deep SADOOD level=Input, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 97.19 | 17.37 | |
| AP-OODOOD level=Output, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 97.14 | 19.52 | |
| AP-OODBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Input, Unsupervised=true2026.02 | 95.61 | 21.85 | |
| Binary logitsOOD level=Input, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 95.42 | 20.51 | |
| KNNBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Output, Unsupervised=true2026.02 | 95.05 | 31.95 | |
| AP-OODBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Output, Unsupervised=true2026.02 | 94.44 | 26.42 | |
| Binary logitsOOD level=Output, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 94.4 | 28.68 | |
| KNNBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Input, Unsupervised=true2026.02 | 94.25 | 35 | |
| MahalanobisBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Input, Unsupervised=true2026.02 | 93.28 | 37.8 | |
| Deep SVDDBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Input, Unsupervised=true2026.02 | 92.15 | 46.28 | |
| Deep SADOOD level=Output, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 91.49 | 51.8 | |
| PerplexityBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Output, Unsupervised=true2026.02 | 91.34 | 44.94 | |
| MahalanobisBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Output, Unsupervised=true2026.02 | 91.07 | 54.49 | |
| Deep SVDDBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Output, Unsupervised=true2026.02 | 89.57 | 57.89 | |
| Relative MahalanobisOOD level=Input, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 88.21 | 29.06 | |
| Relative MahalanobisOOD level=Output, Learning paradigm=Supervised, Backbone=Transformer (base)2026.02 | 87.68 | 37.33 | |
| EntropyBackbone=Transformer (base), Training Data=WMT15 En-Fr, Evaluation Side=Output, Unsupervised=true2026.02 | 74.86 | 68.69 |