Exploratory Landscape Analysis Based Parameter Control

Parameter tuning in evolutionary algorithms is a very important topic, as the correct choice of parameters greatly affects their performance. Fitness landscape analysis can help identify similar problems and allow for gathering problem structure insights for fitness-aware optimization algorithm parameter choice. In this paper, we present an approach to an automatic dynamic parameter control method that uses exploratory landscape analysis and machine learning. Using a dataset of optimal parameter values we collected on different instances of W-model benchmark problem, we trained a machine learning model capable of suggesting parameter values for the (1 + (λ, λ)) genetic algorithm. The results of our experiments show that the machine learning model is able to capture important landscape features and recommend algorithm parameters based on this information. The comparison results with other tuning methods suggest this approach is more effective than static tuning or heuristics-based dynamic parameter control.