Reliability Classification on AIMNet2 held-out molecules 3.76M (test)
71.6Accuracy (all)PROBE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PROBE2026.05 | 71.6 | 83.2 | 93.2 | |
| Majority-class baselineprediction_strategy=always predicts reliable2026.05 | 60 | — | — | |
| Ensemble stdnumber of models=4, scaling=1/√Natoms, threshold=median2026.05 | 57.6 | — | — |