A Clustering Based Niching Method for Evolutionary Algorithms

Abstract

We propose the Clustering Based Niching (CBN) method for Evolutionary Algorithms (EA) to identify multiple global and local optima in a multimodal search space. The basic idea is to apply the biological concept of species in separate ecological niches to EA to preserve diversity. We model species using a multipopulation approach, one population for each species. To identify species in a EA population we apply a clustering algorithm based on the most suitable individual geno-/phenotype representation. One of our goals is to make the niching method as independent of the underlying EA method as possible in such a way that it can be applied to multiple EA methods and that the impact of the niching method on the EA mechanism is as small as possible. CBN starts with a single primordial unclustered population P 0 . Then the CBN-EA generational cycle is entered. First for each population P i one complete EA generation of evaluation, selection and reproduction is simulated. Now CBN starts with the differentiation of the populations by calling the clustering algorithm on each P i . If multiple clusters are found in P i , it splits into multiple new populations. All individuals of P i not included in the clusters found are moved to P 0 as straying loners. To prevent multiple populations to explore the same niche CBN uses representatives (e.g. a centroid) of all populations P i>0 to determine if populations are to be merged. To stabilize the results of the clustering algorithm we currently reduce the mutation step size within all clustered populations P i>0 . A detailed description of the CBN model can be found in [2]. Of course the performance of CBN depends on the clustering algorithm used, since this algorithm specifies the number and kind of niches that can be distinguished. We decided to use the density-based clustering [1] which can identify an a priori unknown number of niches of arbitrary size, shape and spacing. This multi-population approach of CBN replaces the global selection of a standard EA with localized niche based selection and mating. This ensures the survival of each identified niche if necessary. Also each converged population P i>0 directly designates a local/global optimum.

Regulatory Motif Discovery Using a Population Clustering Evolutionary Algorithm

“…In this work, we use population clustering [28], [32] to maintain solution diversity. Population clustering uses a data clustering algorithm to partition the population into subpopulations prior to mating.…”

Learnheuristics: hybridizing metaheuristics with machine learning for optimization with dynamic inputs

“…The most common technique for promoting diversity is clustering analysis. In [70], for instance, individuals in a GA are separated in different subpopulations based on their features and only those in the same cluster compete for survival. The selection operator is applied independently to each cluster.…”

A co-evolutionary framework for regulatory motif discovery

“…Unlike distributed populations [2], which also have explicit partitioning, this partitioning is determined by similarity between solutions rather than by evolutionary history, promoting better coverage of the search space. There have been several other examples of using population clustering within evolutionary algorithms, including [22] and [25].…”