Diversity-Guided Evolutionary Algorithms

Using evolution to improve neural network learning: pitfalls and solutions

“…This indicated that these populations did indeed constitute a true local maximum of fitness, rather than a quirk of the SS approach. Normally, such premature convergence of evolutionary populations to local optima indicates that the parameter space is not being explored effectively, either because of a loss of genetic diversity, or because sufficient diversity was not there in the first place [24].…”

Simulated Quenching Algorithm for Frequency Planning in Cellular Systems

“…This diversity is a crucial issue in the performance of any evolutionary algorithm, including GAs: standard GAs have a tendency to converge prematurely to local optima, mainly due to selection pressure and too high gene flow between population members (Ursem, 2002). A high selection pressure will fill the population with clones of the fittest individuals and it may result in convergence to local minima.…”

Heuristic algorithms and learning techniques: applications to the graph coloring problem

“…In diversity-guided or diversity-controlling genetic algorithms [Ursem, 2002;Shimodaira, 1997;Zhu, 2003], one uses an indicator of overall population diversity so as to choose the genetic operators and their application probability. For example, [Zhu, 2003] uses a diversity indicator based on the average hamming distance to adapt the mutation and crossover rates.…”

“…For example, [Zhu, 2003] uses a diversity indicator based on the average hamming distance to adapt the mutation and crossover rates. In [Ursem, 2002], one employs intensification operators (selection and recombination) when a diversity indicator is high, or diversification operators (mutation) when the diversity is too low. In this research thread, one does not really need a distance measure between individuals, but only an overall population diversity indicator.…”

“…In this research thread, one does not really need a distance measure between individuals, but only an overall population diversity indicator. For example, [Ursem, 2002] employs an indicator of statistical dispersion around the average value on each variable, making it particularly useful in continuous optimization. Generally speaking, the population diversity can be measured by numerous indicators, depending on the field (see the refs in [Zhu, 2003, p. 1…”