In almost every major field of evolutionary computing, population size is one of the parameters that
a researcher has to deal with. In nature, natural selection as the driving force of evolution acts upon
populations, and in computing science, group search is arguably the unique feature of evolutionary
computing. Therefore, it was no surprise that the pioneers of evolutionary computing, such as De
Jong (1975) and Holland (1975), had already studied the population sizing from the very beginning
of evolutionary computing. What is perhaps surprising is that more than three decades later, we still
largely depend on the experience or ad-hoc trial-and-error approach to set the population size. In
their recent monograph for evolutionary computing, Eiben and Smith (2003) indicated: "In almost
all EC applications, the population size is constant and does not change during the evolutionary
search". Does the status quo imply that population size is not an important parameter?
The answer to the previous question seems a definite no. Current practice of manual setting of
population size in evolutionary computation is experience-based, but not robust. Too small of
populations can lead to premature convergence, and too big of populations can be computationally
costly. In particular, in Genetic programming, a big population is often a culprit for the too early
emergence of code bloat, which may cause failure or even crash the system. This manual setting by
experienced programmers is acceptable for design modeling or ad-hoc applications; it is problematic
for real-time applications, or in the fields that require more predictability and robustness. In the
latter scenarios, automatic, adaptive and robust population sizing (also other parameters such as
mutation and crossover rates) is necessary. On the practical side, what is equally important is the
robustness and adaptability of the algorithms. Furthermore, many of the problems we face are NPhard
problems; efficient population sizing may have significant impacts on the success of the
heuristic algorithms. However, an optimal population size effective in exploring fitness space and
efficient in utilizing computing resources is theoretically very intriguing.
Much of the evolutionary computation has been inspired by genetics-centered evolution theory. The
terminology in evolutionary computing such as gene, chromosome, genotype, phenotype, fitness
landscape, genetic algorithms, and genetic programming are the clear indication of the influences of
genetics, and the fixed population size itself has clear traces in quantitative or theoretical genetics
(e.g., Felsenstein 2007). For example, fixed population or infinite population size is simply more
convenient to model the gene frequency theoretically, and one example of infinite population size is
the Hardy-Weinberg law.
In nature, population size is neither fixed nor infinite, instead ecological population is always
dynamic with complex self-regulation and constrained by environment. It is the population
individuals or phenotypes, rather than genotypes, that directly interact with environment. The study
2
of interactions between phenotypes and environment belongs to ecology, and population dynamics
has been the central focus of population ecology. Therefore, in this study, we emphasize the
ecological dimension in evolutionary computation.
Therefore, we argue that it is the time to bring back the ecological dimension into evolutionary
computation. We will try to demonstrate that, at least in the case of population-sizing, the
population dynamics and evolutionary game theory provide a better alternative approach than the
traditional genetics-centered evolution paradigm. We further suggest that it should be helpful for
whole evolutionary computation to pay more active attention to ecology. The relationship between
ecology and evolution is best described with G. E. Hutchinson's treatise title, The Ecological Theater
and the Evolutionary Play. Anyway, without ecological theater, natural selection has no place to act and
evolution cannot happen.
This paper is a mixture of the original research and a comprehensive review of the state-of-the-art
studies from a multiple-discipline perceptive. Thus, it not only reveals the existing problems, but
also summarizes (in Section 6) the approaches the authors developed to remedy or solve the
problems. Some of the approaches in Section 6 have been published and the others are under peerreview
processes.