Table 2 Solution accuracy for citation graphs in terms of wrongly colored edges (lower value is better) comparison of heuristic (Tabucol), learning-based approaches (GNN and PI-SAGE), Ising methods, and Vectorized framework
From: Efficient optimization accelerator framework for multi-state spin Ising problems
Problem | #nodes | #edges | #colors | #GNN33 | #PI-SAGE33 | #Tabucol | #Probabilistic ising | #Simulated bifurcation | #Vectorized GPU | # Probabilistic ising + Parallel Tempering GPU | #Vectorized+ Parallel Tempering GPU |
|---|---|---|---|---|---|---|---|---|---|---|---|
cora | 2708 | 5278 | 5 | 3 | 0 | 0 | 1004 | 486 | 2 | 830 | 1 |
citseer | 3279 | 4552 | 6 | 3 | 0 | 0 | 707 | 163 | 1 | 592 | 0 |
pubmed | 19717 | 44324 | 8 | 35 | 17 | 17 | NA | NA | 18 | NA | 11 |
