Niching in Evolution Strategies: An Introduction (Last Update: 2012)

Ofer M. Shir, PhD

Evolutionary Algorithms (EAs), popular population-based stochastic search-methods, have the tendency to lose diversity within their population of feasible solutions and to converge into a single solution. Niching methods, the extension of EAs to multi-modal optimization, address this issue by maintaining the diversity of certain properties within the population - and this way they allow parallel convergence into multiple good solutions in multimodal domains. To this end, niching methods have been studied mainly within the field of Genetic Algorithms (GAs). The research in this direction has yielded various successful methods which have been shown to find multiple solutions efficiently, but naturally were limited to low-dimensional real-valued problems. Evolution Strategies (ES) are a canonical EA for real-valued function optimization, due to their straightforward encoding, their specific variation operators, the self-adaptation of their mutation distribution as well as to their high performance in this domain in comparison with other methods on benchmark problems. The higher the dimensionality of the search space, the more suitable a task becomes for an ES.

The study of niching is challenging both from the theoretical point of view and from the practical point of view. The theoretical challenge is two-fold - maintaining the diversity within a population-based stochastic algorithm from the computational perspective, but also having an insight into speciation theory from the biological perspective. The practical aspect provides a real-world motivation for this problem - there is an increasing interest of the applications' community in providing the decision maker with multiple solutions with different conceptual designs, for single- or multi-objective search spaces.

                      

[LEFT] An analogy: find the best runners with the highest genetic diversity among each other.  [CENTER] Natural speciation: illustration.  [RIGHT] Speciation table: butterflies.

Literature

Niching in Evolutionary Algorithms

Evolution Strategies

Derandomized Evolution Strategies

Classical Niching Concepts

Niching in Evolution Strategies

The Niche Radius Problem

Miscellaneous



Illustration: Niching in Action