Uncertainty Estimation on NQ-open
75.88AUROCSE+MCH
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SE+MCHBase Model=Llama2-chat, Retriever=Contriver, Calibration=MCH2025.05 | 75.88 | 88.02 | 32.52 | 91.53 | 46.9 | |
| EigV+MCHBase Model=Llama2-chat, Retriever=Contriver, Calibration=MCH2025.05 | 73.6 | 88.67 | 23.93 | 93.82 | 34.88 | |
| ECC+MCHBase Model=Llama2-chat, Retriever=Contriver, Calibration=MCH2025.05 | 71.81 | 86.08 | 26.99 | 89.93 | 39.54 | |
| PE+MCHBase Model=Llama2-chat, Retriever=Contriver, Calibration=MCH2025.05 | 69.85 | 76.05 | 37.42 | 83.75 | 51.93 | |
| TotalBackbone=Mistral-7B2026.04 | 69.8 | — | — | — | — | |
| SE+NLIBase Model=Llama2-chat, Retriever=Contriver, Calibration=NLI2025.05 | 69.77 | 79.61 | 40.49 | 86.72 | 50 | |
| EigV+NLIBase Model=Llama2-chat, Retriever=Contriver, Calibration=NLI2025.05 | 69.44 | 80.91 | 27.61 | 91.76 | 35.27 | |
| AleatoricBackbone=Mistral-7B2026.04 | 69 | — | — | — | — | |
| ECC+NLIBase Model=Llama2-chat, Retriever=Contriver, Calibration=NLI2025.05 | 68.96 | 78.64 | 32.52 | 87.18 | 42.64 | |
| SelfCheckGPTBackbone=Mistral-7B2026.04 | 68.8 | — | — | — | — | |
| SEBase Model=Llama2-chat, Retriever=Contriver, Calibration=None2025.05 | 67.49 | 77.35 | 33.75 | 91.53 | 36.05 | |
| Mean Token EntropyBackbone=Mistral-7B2026.04 | 67.4 | — | — | — | — | |
| ECC+CTIBase Model=Llama2-chat, Retriever=Contriver, Calibration=CTI2025.05 | 67.37 | 75.73 | 23.31 | 87.18 | 37.98 | |
| PE+NLIBase Model=Llama2-chat, Retriever=Contriver, Calibration=NLI2025.05 | 67.21 | 66.02 | 47.24 | 77.57 | 55.43 | |
| Token EntropyBackbone=Mistral-7B2026.04 | 67.2 | — | — | — | — | |
| SE+CTIBase Model=Llama2-chat, Retriever=Contriver, Calibration=CTI2025.05 | 67.09 | 77.02 | 25.76 | 89.47 | 40.31 | |
| EigV+CTIBase Model=Llama2-chat, Retriever=Contriver, Calibration=CTI2025.05 | 66.82 | 77.35 | 20.25 | 90.16 | 34.5 | |
| SemanticEntropyBackbone=Mistral-7B2026.04 | 65.6 | — | — | — | — | |
| PE+CTIBase Model=Llama2-chat, Retriever=Contriver, Calibration=CTI2025.05 | 65.38 | 61.49 | 44.17 | 76.43 | 53.88 | |
| PerplexityBackbone=Mistral-7B2026.04 | 65.2 | — | — | — | — | |
| Max Token Prob.Backbone=Mistral-7B2026.04 | 65.2 | — | — | — | — | |
| PEBase Model=Llama2-chat, Retriever=Contriver, Calibration=None2025.05 | 64.87 | 60.19 | 47.85 | 77.35 | 51.16 | |
| SC + VCBackbone=Mistral-7B2026.04 | 64.1 | — | — | — | — | |
| Max Sequence Prob.Backbone=Mistral-7B2026.04 | 64 | — | — | — | — | |
| EigVBase Model=Llama2-chat, Retriever=Contriver, Calibration=None2025.05 | 63.94 | 65.37 | 12.88 | 88.56 | 24.42 | |
| ECCBase Model=Llama2-chat, Retriever=Contriver, Calibration=None2025.05 | 63.57 | 61.49 | 9.82 | 83.06 | 18.99 | |
| SC Based VCBackbone=Mistral-7B2026.04 | 63 | — | — | — | — | |
| SC ScoreBackbone=Mistral-7B2026.04 | 61.6 | — | — | — | — | |
| Self CertaintyBackbone=Mistral-7B2026.04 | 59.8 | — | — | — | — | |
| Closeness CentralityBackbone=Mistral-7B2026.04 | 59.7 | — | — | — | — | |
| Kernel Lang. Ent.Backbone=Mistral-7B2026.04 | 58.5 | — | — | — | — | |
| PTrueBackbone=Mistral-7B2026.04 | 55 | — | — | — | — |