Out-of-distribution detection on TREC (test)
0.8FPR@90MSP + OE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MSP + OEArchitecture=2-layer GRU, Training=Fine-tuned with Outlier Exposure2018.12 | 0.8 | 99.3 | 97.6 | — | |
| AD-KDTeacher training dataset=Banking77 (IND)2026.02 | 7.2 | 97.3 | — | 19.2 | |
| CUDTeacher training dataset=Banking77 (IND)2026.02 | 8.6 | 97.45 | — | 16.6 | |
| CKDTeacher training dataset=Banking77 (IND)2026.02 | 15 | 95.18 | — | 31.4 | |
| TinyBERTTeacher training dataset=Banking77 (IND)2026.02 | 42.6 | 82.34 | — | 61.8 | |
| MSPArchitecture=2-layer GRU, Training=Standard2018.12 | 43.5 | 82.1 | 52.2 | — | |
| OECCBackbone=2-layer GRUs, Fine-tuning Protocol=OECC, Fine-tuning Epochs=2, Batch size=64, Learning rate=0.012019.06 | 75 | 99.32 | 97.52 | — | |
| OEBackbone=2-layer GRUs, Fine-tuning Protocol=Outlier Exposure, Fine-tuning Epochs=2, Batch size=64, Learning rate=0.012019.06 | 78 | 99.28 | 97.64 | — |