ABSTRACT
Evolution Strategies and some continuous domain Estimation of Distribution Algorithms are stochastic optimization procedures based on sampling multivariate Gaussian (normal) distributions. They can be formulated in a common, unified, but still very simple framework. Such a framework is very useful to understand subtle differences of algorithms.
This tutorial focuses on the most important question: how to chose and update the sample distribution parameters. The most common and successful approaches are reviewed. Covered methods include self-adaptation, success rules, path length control, Covariance Matrix Adaptation CMA), and Estimation of Multivariate Normal Algorithm (EMNA).
Methods are motivated with respect to the difficulties one has to face when solving continuous domain non-linear, non-convex optimization problems. Specific problem difficulties will be discussed, for example ill-conditioning and non-separability.
Index Terms
- Evolution strategies and related estimation of distribution algorithms
Recommendations
Adaptive parent population sizing in evolution strategies
Adaptive population sizing aims at improving the overall progress of an evolution strategy. At each generation, it determines the parental population size that promises the largest fitness gain, based on the information collected during the evolutionary ...
Reconsidering constraint release for active-set evolution strategies
GECCO '17: Proceedings of the Genetic and Evolutionary Computation ConferenceWe consider the problem of solving constrained numerical optimization problems where the objective function is a black box, but the constraint functions are known explicitly. A recently proposed active-set approach implemented in an evolution strategy ...
Learning probability distributions in continuous evolutionary algorithms– a comparative review
We present a comparative review of Evolutionary Algorithms that generate new population members by sampling a probability distribution constructed during the optimization process. We present a unifying formulation for five such algorithms that enables us ...
Comments