Graph Algorithmic Reasoning on GAR Cycle Detection
98.07AccuracyGITA
Evaluation results
| Method | Links | |
|---|---|---|
| GITATraining Paradigm=Task-Specific Training2026.01 | 98.07 | |
| G-SubstrateTraining Paradigm=Multi-Task Training2026.01 | 96.97 | |
| Unified multi-taskTraining Paradigm=Multi-Task Training2026.01 | 96.19 | |
| G-WizTraining Paradigm=Task-Specific Training2026.01 | 95.46 | |
| Naive multi-taskTraining Paradigm=Multi-Task Training2026.01 | 94.72 | |
| Unified single-taskTraining Paradigm=Task-Specific Training2026.01 | 94.7 | |
| Naive multi-task + interleaveTraining Paradigm=Multi-Task Training2026.01 | 93.86 | |
| Naive single-taskTraining Paradigm=Task-Specific Training2026.01 | 92.18 |