Calibration on Four graduate-level benchmarks clean records
7.5ECEMax Probability
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Max Probability2026.02 | 7.5 | 71.9 | 2 | |
| Margin2026.02 | 9.5 | 71.6 | 2.7 | |
| Entropy2026.02 | 20.9 | 68.4 | 7.6 | |
| αencoding parameter (s)=0.92026.02 | 38.7 | — | 16.8 |