A Visualizable Test Problem Generator for Many-Objective Optimization

Visualizing the search behavior of a series points or populations in their native domain is critical understanding biases and attractors an optimization process. Distance-based many-objective test problems have been developed to facilitate visualization 2-D design space with arbitrarily many objective functions. Previous works proposed few commonly seen problem characteristics into this framework, such as definition disconnected Pareto sets dominance resistant regions space. The authors’ previous work has advanced research further by providing generator automatically create user-defined instances featuring any combination these features well newly introduced ones, landscape discontinuities, varying ranges, neutrality. This makes number additional contributions including proposal enhanced, open-source feature-rich that can exhibiting range features—some which are here form extensions existing features. A comprehensive validation also provided using popular multiobjective algorithms, some settings different challenges for optimizer identified.