Token-level Uncertainty Quantification on MIMIC-DI
89.11AUCROCReverse Probing (XGBoost)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Reverse Probing (XGBoost)Backbone LLM=BioMistral-7B2026.05 | 89.11 | 54.78 | |
| Reverse Probing (CatBoost)Backbone LLM=OpenBioLLM-8B2026.05 | 88.48 | 50.29 | |
| Reverse Probing (XGBoost)Backbone LLM=OpenBioLLM-8B2026.05 | 88.33 | 49.76 | |
| Reverse Probing (CatBoost)Backbone LLM=BioMistral-7B2026.05 | 86.91 | 51.78 | |
| SE ProbesBackbone LLM=BioMistral-7B2026.05 | 72.04 | 13.13 | |
| Semantic EntropyBackbone LLM=BioMistral-7B2026.05 | 64.48 | 8.68 | |
| SE ProbesBackbone LLM=OpenBioLLM-8B2026.05 | 60.15 | 5.87 | |
| Decomposing UncertaintyBackbone LLM=BioMistral-7B2026.05 | 59.38 | 8.25 | |
| Sliding Window EntropyBackbone LLM=OpenBioLLM-8B2026.05 | 58.77 | 12.67 | |
| Semantic EntropyBackbone LLM=OpenBioLLM-8B2026.05 | 58.04 | 8.54 | |
| Decomposing UncertaintyBackbone LLM=OpenBioLLM-8B2026.05 | 56.57 | 7.75 | |
| Uncertainty Aware HeadsBackbone LLM=BioMistral-7B2026.05 | 55.08 | 8.1 | |
| Uncertainty Aware HeadsBackbone LLM=OpenBioLLM-8B2026.05 | 54.88 | 8.52 | |
| Semantic EnergyBackbone LLM=OpenBioLLM-8B2026.05 | 53.75 | 9.14 | |
| Attention ChainBackbone LLM=BioMistral-7B2026.05 | 51.95 | 8.74 | |
| Sliding Window EntropyBackbone LLM=BioMistral-7B2026.05 | 51.27 | 8.99 | |
| Attention ChainBackbone LLM=OpenBioLLM-8B2026.05 | 49.13 | 10.12 | |
| Token EntropyBackbone LLM=OpenBioLLM-8B2026.05 | 48.32 | 7.67 | |
| Semantic EnergyBackbone LLM=BioMistral-7B2026.05 | 47.14 | 8.1 | |
| Token EntropyBackbone LLM=BioMistral-7B2026.05 | 43.37 | 6.58 |