A Graph Neural Network for pairwise surrogate modeling in population-based algorithms with tournament selection

Abstract

Optimization problems widely arise in various science and engineering fields. Optimisation involves evaluating a candidate solution, which can be computationally intensive. Machine learning-based surrogate models can contribute to learning the specific pattern among the decision variables and objective values to reduce the computation time of fitness evaluation. In this study, we have proposed a novel pairwise surrogate model to identify the superiority between candidate solutions in a pairwise comparison. We demonstrated a Graph Neural Network (GNN) to be trained on number of pairs, then utilized to compare a pair of candidate solutions. To examine the efficacy of our model, we utilized the surrogate model on CEC2017 benchmarks in different dimensions. Moreover, the result of surrogate-assisted and none-assisted form of two well-known optimization algorithms were compared. Results show that the proposed method can significantly reduce the computing cost. In the presence of higher dimensions, our model is more effective than most surrogate models for comparison-based optimizers.