2542
Comment:

← Revision 5 as of 20180321 08:39:50 ⇥
2672

Deletions are marked like this.  Additions are marked like this. 
Line 1:  Line 1: 
In computer science, an '''evolution strategy (ES)''' is an [[optimization]] technique based on ideas of [[evolution]]. It belongs to the general class of [[evolutionary computation]] or [[artificial evolution]] methodologies.<<FootNote(“Evolution Strategy” ''WikiPedia'', 19 March. 2018, [[https://en.wikipedia.org/wiki/Evolution_strategyhttps://en.wikipedia.org/wiki/Evolution_strategy]])>>  In computer science, an '''evolution strategy (ES)''' is an [[Optimizationoptimization]] technique based on ideas of [[evolution]]. It belongs to the general class of [[Evolutionary Computationevolutionary computation]] or [[Artificial Evolutionartificial evolution]] methodologies.<<FootNote(“Evolution Strategy” ''WikiPedia'', 19 March. 2018, [[https://en.wikipedia.org/wiki/Evolution_strategyhttps://en.wikipedia.org/wiki/Evolution_strategy]])>> 
Line 9:  Line 9: 
Evolution strategies use natural problemdependent representations, and primarily [[mutation]] and [[Selection]], as search operators. In common with [[evolutionary algorithms]], the operators are applied in a loop. An iteration of the loop is called a generation. The sequence of generations is continued until a termination criterion is met.  Evolution strategies use natural problemdependent representations, and primarily [[Mutationmutation]] and [[Selectionselection]], as search operators. In common with [[Evolutionary Algorithmsevolutionary algorithms]], the operators are applied in a loop. An iteration of the loop is called a generation. The sequence of generations is continued until a termination criterion is met. 
Line 11:  Line 11: 
As far as realvalued search spaces are concerned, mutation is normally performed by adding a [[normal distributionnormally distributed]] random value to each vector component. The step size or mutation strength (i.e. the standard deviation of the normal distribution) is often governed by selfadaptation (see [[evolution window]]). Individual step sizes for each coordinate or correlations between coordinates are either governed by selfadaptation or by covariance matrix adaptation ([[CMAES]]).  As far as realvalued search spaces are concerned, mutation is normally performed by adding a [[Normal Distributionnormally distributed]] random value to each vector component. The step size or mutation strength (i.e. the standard deviation of the normal distribution) is often governed by selfadaptation (see [[Evolution Windowevolution window]]). Individual step sizes for each coordinate or correlations between coordinates are either governed by selfadaptation or by covariance matrix adaptation ([[CMAES]]). 
Line 13:  Line 13: 
The (environmental) selection in evolution strategies is deterministic and only based on the fitness rankings, not on the actual fitness values. The resulting algorithm is therefore invariant with respect to monotonic transformations of the objective function. The simplest evolution strategy operates on a population of size two: the current point (parent) and the result of its mutation. Only if the mutant's fitness is at least as good as the parent one, it becomes the parent of the next generation. Otherwise the mutant is disregarded. This is a ''(1 + 1)ES''. More generally, λ mutants can be generated and compete with the parent, called ''(1 + λ)ES''. In (1 , λ)ES the best mutant becomes the parent of the next generation while the current parent is always disregarded. For some of these variants, proofs of [[Rate of convergencelinear convergence]] (in a [[stochastic]] sense) have been derived on unimodal objective functions.  The (environmental) selection in evolution strategies is deterministic and only based on the fitness rankings, not on the actual fitness values. The resulting algorithm is therefore invariant with respect to monotonic transformations of the objective function. The simplest evolution strategy operates on a population of size two: the current point (parent) and the result of its mutation. Only if the mutant's fitness is at least as good as the parent one, it becomes the parent of the next generation. Otherwise the mutant is disregarded. This is a ''(1 + 1)ES''. More generally, λ mutants can be generated and compete with the parent, called ''(1 + λ)ES''. In (1 , λ)ES the best mutant becomes the parent of the next generation while the current parent is always disregarded. For some of these variants, proofs of [[Rate of convergencelinear convergence]] (in a [[Stochasticstochastic]] sense) have been derived on unimodal objective functions. 
In computer science, an evolution strategy (ES) is an optimization technique based on ideas of evolution. It belongs to the general class of evolutionary computation or artificial evolution methodologies.^{1}
History
The 'evolution strategy' optimization technique was created in the early 1960s and developed further in the 1970s and later by Ingo Rechenberg, HansPaul Schwefel and their coworkers.
Methods
Evolution strategies use natural problemdependent representations, and primarily mutation and selection, as search operators. In common with evolutionary algorithms, the operators are applied in a loop. An iteration of the loop is called a generation. The sequence of generations is continued until a termination criterion is met.
As far as realvalued search spaces are concerned, mutation is normally performed by adding a normally distributed random value to each vector component. The step size or mutation strength (i.e. the standard deviation of the normal distribution) is often governed by selfadaptation (see evolution window). Individual step sizes for each coordinate or correlations between coordinates are either governed by selfadaptation or by covariance matrix adaptation (CMAES).
The (environmental) selection in evolution strategies is deterministic and only based on the fitness rankings, not on the actual fitness values. The resulting algorithm is therefore invariant with respect to monotonic transformations of the objective function. The simplest evolution strategy operates on a population of size two: the current point (parent) and the result of its mutation. Only if the mutant's fitness is at least as good as the parent one, it becomes the parent of the next generation. Otherwise the mutant is disregarded. This is a (1 + 1)ES. More generally, λ mutants can be generated and compete with the parent, called (1 + λ)ES. In (1 , λ)ES the best mutant becomes the parent of the next generation while the current parent is always disregarded. For some of these variants, proofs of linear convergence (in a stochastic sense) have been derived on unimodal objective functions.
Sources
Note: first version of this page is based entirely on the first citation
“Evolution Strategy” WikiPedia, 19 March. 2018, https://en.wikipedia.org/wiki/Evolution_strategy (1)