Out-of-Distribution Detection on Snips CLINC IND OOD
2.78EERDRM+L-Mahalanobis
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DRM+L-MahalanobisBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 2.78 | 2.78 | 3.16 | 98.53 | 99.76 | 87.02 | |
| DRM+L-MahalanobisBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 3 | 1.95 | 2.63 | 98.9 | 98.15 | 99.24 | |
| En-AEBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 3.59 | 3.08 | 3.64 | 88.96 | 94.39 | 86.84 | |
| MahalanobisBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 3.9 | 2.66 | 3.47 | 98.79 | 97.73 | 99.21 | |
| En-AEBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 4.4 | 4.18 | 4.25 | 98.56 | 97.41 | 98.12 | |
| G-ODINBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 9.65 | 13.31 | 8.32 | 97.21 | 95.7 | 98.02 | |
| ConfidenceBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 9.91 | 14.94 | 9.18 | 96.09 | 94.78 | 97.21 | |
| ODINBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 10.01 | 14.22 | 9.43 | 96.46 | 94.59 | 97.75 | |
| EntropyBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 10.21 | 14.54 | 9.25 | 96.32 | 94.9 | 97.53 | |
| MahalanobisBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 11.11 | 5.58 | 10.21 | 95.61 | 99.22 | 76.61 | |
| G-ODINBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 22.02 | 55.67 | 21.82 | 85.6 | 97.73 | 50.38 | |
| ConfidenceBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 22.22 | 51.85 | 19.34 | 87.44 | 97.67 | 55.16 | |
| EntropyBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 23.15 | 52.68 | 19.58 | 87.12 | 97.6 | 52.27 | |
| ODINBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 23.15 | 58.33 | 23.01 | 83.58 | 96.63 | 47.36 | |
| En-ConGANBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 54.22 | 99.1 | 37.32 | 45.64 | 30.03 | 52.59 | |
| En-ConGANBackbone=RoBERTa, Evaluation Protocol=Model Finetuning2021.06 | 54.5 | 99.16 | 42.61 | 39.03 | 37.15 | 51.23 |