Uncertainty Estimation on SQW (SearchQA)
0.83AUROCProposed
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ProposedBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.83 | 0.91 | |
| Geometry-Calibrated Conformal AbstentionBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.81 | 0.86 | |
| SARBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.77 | 0.81 | |
| SemanticBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.77 | 0.82 | |
| PerplexityBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.76 | 0.81 | |
| EigenBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.76 | 0.8 | |
| SARBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.74 | 0.82 | |
| SemanticBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.74 | 0.83 | |
| ATRMDBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.71 | 0.83 | |
| PerplexityBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.7 | 0.8 | |
| EigenBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.7 | 0.8 | |
| FocusBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.55 | 0.67 | |
| ATRMDBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.53 | 0.59 | |
| P(True)Backbone=LLaMA-3.2-3B-Instruct2026.04 | 0.49 | 0.55 | |
| FocusBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.49 | 0.69 | |
| AttentionBackbone Model=LLaMA-3-8B-Instruct2026.04 | 0.47 | 0.63 | |
| P(True)Backbone Model=LLaMA-3-8B-Instruct2026.04 | 0.47 | 0.64 | |
| AttentionBackbone=LLaMA-3.2-3B-Instruct2026.04 | 0.45 | 0.54 |