View source for Talk:Evolution strategies

=================================================
3rd reviewer:

Very good introduction to ES.  The author gives a short and complete
overview of the topic. 

Here some minor comments and suggestions: 

* The <math>(\mu/\mu_I, \lambda)-\sigma</math>-Self-Adaptation-ES does not adapt
the optimal step sizes, in the static case, for <math>\mu>1</math> (in dfference to  cumulated step size adaptation). Since this  self-adaptation-scheme is 
very simple, the author should not encourage the reader to use it for <math>\mu>1</math>. 
Perhaps he can change the title to <math>(1, \lambda)-\sigma</math>-Self-Adaptation-ES.

* In the Matlab-implementation of (\mu/\mu_I, \lambda)-\sigma-Self-Adaptation_ES:
The line gives errors in Matlab
  OffspringIndividual.y = Recombinant.y + OffspringIndividual.sigma * randn(1,n); % mutate object parameter
it should be
  OffspringIndividual.y = Recombinant.y + OffspringIndividual.sigma * randn(n,1); % mutate object parameter

* The performance of an ES on a specific problem class depends also crucially on the design 
of the adaptation scheme.

* Not only the variation operators should respect the principle of strong causality. 
The goal function, the representation (encoding) and the the variation operators
should respect the principle of strong causality. For many real valued optimizaton
problems it is easier to change the representation than the variation operators.

* It should be stated that a simplified for of the CMA-ES is presented.

* Perhaps the author can include some statements about the time complexity of an ES. 

* I have corrected the links, thus they lead to the actual webpages