Machine learning-assisted generation of unexplored solutions in a targeted region of the Pareto front for multi-objective optimization

Multi-objective optimization problems aim to identify solutions that maximize or minimize conflicting objectives. Population-based multi-objective algorithms are effective in exploring the search space but often provide limited solutions within the decision-makers' region of interest on the Pareto front. Recent advances in machine learning have shown promise in generating solutions; however, they suffer from a lack of control and require knowledge of the objective function's attributes. This study proposes a framework using machine learning models to generate unexplored solutions in the targeted region of the Pareto front and alleviate its sparsity. By employing diverse sampling techniques and integrating long-term memory within an optimization process, the framework can produce more than twice as many solutions in the targeted region. As demonstrated in experiments with eight real-world problems and thirteen benchmark problems from the MaF test suite, with various numbers of objectives, the proposed method can facilitate decision-making by providing many innovative solutions as a post-optimization process.