Towards Benchmarking Multi-Objective Optimization Algorithms Based on the Basin Connectivity

When applying evolutionary computation for multimodal multiobjective optimization, it is crucial to balance exploration and exploitation: exploring the design space to identify regions that contain Pareto optima and exploiting these regions to converge towards the Pareto optima. When it comes to benchmarking for multi-objective optimization, however, it has been difficult to ensure such a landscape with desired properties. We thus propose to pre-determine the landscape and implement it as a benchmark problem. This framework is named the Benchmarking Based on Basin Connectivity (3BC) and targets continuous optimization. The 3BC suite includes an instance with recursively nested basins, forming a funnel structure, and another instance without nesting. Our implementation is publicly available at https://github.com/dsakurai/benchmark-visualizer.