Calibration on GSM8K
1.64ECECAGE-CAL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CAGE-CALType=Trained2026.05 | 1.64 | 84.04 | |
| Plurality share + Scaling-bin.Type=Post-hoc plurality, Technique=Scaling-binning2026.05 | 2.28 | 82.85 | |
| DiscoUQ-LLMType=Trained2026.05 | 2.43 | 83.19 | |
| Plurality share + PlattType=Post-hoc plurality, Technique=Platt scaling2026.05 | 2.86 | 83.01 | |
| Plurality share + IsotonicType=Post-hoc plurality, Technique=Isotonic regression2026.05 | 3.22 | 80 | |
| Scalar + GBTType=Trained2026.05 | 4.34 | 76.35 | |
| Collab. Cal.Type=LLM-elicited2026.05 | 8.01 | 70.01 | |
| LLM-Cal (+topo)Type=LLM-elicited, Topology=Included2026.05 | 8.28 | 71.16 | |
| LLM-Cal (no topo)Type=LLM-elicited, Topology=None2026.05 | 8.96 | 67.18 | |
| Plurality shareType=Post-hoc plurality2026.05 | 17.98 | 83.01 | |
| GraphCalType=Trained2026.05 | 19.7 | 81.02 |