Differential Evolution with an Unbounded Population
Tomofumi Kitamura, Alex S. Fukunaga · 2022 IEEE Congress on Evolutionary Computation (CEC) · 2022
The notion of a population with individuals which are replaced by newly generated individuals is a pervasive idea in differential evolution (DE). Techniques such as archives have been previously proposed to supplement the central idea of a population, motivated by improved performance compared to a simple population structure. We consider Unbounded DE (UDE), which uses a population where individuals are never replaced, and the population monotonically grows as new individuals are generated. Behaviors similar to standard populations, as well as previous population modifications such as archives can be implemented within the UDE framework by implementing specific selection policies. We show experimentally that UDE can be configured to be competitive with standard DE as well as adaptive DE algorithms, showing that the traditional notion of replacement is not necessary for good performance.