Graph Algorithmic Reasoning on GAR Connectivity
99.71AccuracyNaive multi-task
Evaluation results
| Method | Links | |
|---|---|---|
| Naive multi-taskTraining Paradigm=Multi-Task Training2026.01 | 99.71 | |
| Naive single-taskTraining Paradigm=Task-Specific Training2026.01 | 99.44 | |
| G-SubstrateTraining Paradigm=Multi-Task Training2026.01 | 98.41 | |
| Naive multi-task + interleaveTraining Paradigm=Multi-Task Training2026.01 | 98.27 | |
| GITATraining Paradigm=Task-Specific Training2026.01 | 98.17 | |
| Unified multi-taskTraining Paradigm=Multi-Task Training2026.01 | 98.09 | |
| Unified single-taskTraining Paradigm=Task-Specific Training2026.01 | 97.8 | |
| G-WizTraining Paradigm=Task-Specific Training2026.01 | 97.74 |