Learning to Solve Combinatorial Optimization Problems on Graphs with State-Aware Multi-Relation Aggregation

Solving a combinatorial optimization problem is a challenging algorithm design task that demands a comprehensive understanding of the tackled problem and the design of a novel strategy to find the optimal solution efficiently. Owing to the advances in machine learning and research interests in exploring, machine learning techniques to tackle combinatorial optimization problems on graphs have grown recently. One of the key challenges in this research effort is to accurately capture the important information in the graph structure and partial solutions appearing in the intermediate steps toward finding the solution. To overcome this issue, we propose a new model, namely, State-Aware Multi-relation Aggregation (SAMA). Experiments conducted on graphs that are artificially generated and appearing in real applications demonstrate the superiority of SAMA over alternative algorithmic and learning-based models.