Success Rate Prediction on MMLU CF
100Perturbation Success RateTRUE (DAG-based predictor)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TRUE (DAG-based predictor)Category=Health and Medicine2026.02 | 100 | 80 | 0.53 | — | 3.98 | |
| Baseline CoT predictorCategory=Health and Medicine2026.02 | 100 | 80 | — | 4.51 | — | |
| TRUE (DAG-based predictor)Category=Economics and Finance2026.02 | 100 | 40 | 1.69 | — | 12.13 | |
| Baseline CoT predictorCategory=Economics and Finance2026.02 | 100 | 40 | — | 13.82 | — | |
| TRUE (DAG-based predictor)Category=Average2026.02 | 86.6 | 65 | 0.97 | — | 3.93 | |
| Baseline CoT predictorCategory=Average2026.02 | 86.6 | 65 | — | 4.9 | — | |
| TRUE (DAG-based predictor)Category=Classification and Rules2026.02 | 83.3 | 65 | 1 | — | -0.17 | |
| Baseline CoT predictorCategory=Classification and Rules2026.02 | 83.3 | 65 | — | 0.83 | — | |
| TRUE (DAG-based predictor)Category=Science and Technology2026.02 | 83.3 | 65 | 0.98 | — | -0.15 | |
| Baseline CoT predictorCategory=Science and Technology2026.02 | 83.3 | 65 | — | 0.83 | — | |
| TRUE (DAG-based predictor)Category=History and Politics2026.02 | 66.6 | 75 | 0.67 | — | 3.84 | |
| Baseline CoT predictorCategory=History and Politics2026.02 | 66.6 | 75 | — | 4.51 | — |