Talk:Evolution strategy

From Wikipedia, the free encyclopedia

Contents

[edit] Original References

One should add the original papers, not just the recent surveys. After all, these are the first serious applications of artificial evolution:

Ingo Rechenberg (1971): Evolutionsstrategie - Optimierung technischer Systeme nach Prinzipien der biologischen Evolution (PhD thesis). Reprinted by Fromman-Holzboog (1973).

Hans-Paul Schwefel (1974): Numerische Optimierung von Computer-Modellen (PhD thesis). Reprinted by by Birkhäuser (1977).

Algorithms 13:13, 2 June 2007 (UTC)

[edit] Use of Bionics?

It appears that the word Bionics is used commonly on the external links to German institutions, but the definition on the wiki and commonly used in English isn't entirely related to the specific use of an algorithm like this?! I've removed the see also link to the page, if anyone has a strong reason why I'm wrong just leave me some stuff on my Talk page :)

"Bionics: Study of the results of biological evolution from the engineering point of view." That's more like biologically inspired engineering which doesn't have a great deal to do with a form of evolutionary algorithm which can be applied to any kind of problem where the solution space is large, it doesn't have to be a biologically inspired problem or solution. Ultima 13:59, 9 November 2006 (UTC)

[edit] Bionics

I guess that would be interesting to keep the bionics link, since the Evolution Strategy was conceived along the Bionics Zeitgeist. According to Professor Ingo Rechenberg, the (1+1)-ES itself can be considered as the maximum abstraction of darwinian evolution. Evolutionary algorithms can be seen as abstractions like that, but from the engineering/computer science point of view. :)

So, let's talk a little more to decide what to do next about the bionics link.

If someone has something to say, please, let me know.

Nosophorus 14:43, 11 November 2006 (UTC)

The general use of the word Bionics is very broad, and although it is used to mean something more specific by external institutions that doesn't mean it should be used in the same way on wikipedia. It's like me going to a page about Horizontal gene transfer and then saying: See also: Biology if you look on that page they don't even link to Evolution they link to more specific articles or articles directly related. But still this is all just pedantry, I have no problem with the link being on there. However, I did remove the link because my supervisor at university took one look at the page and saw the link to Bionics. He straight away thought the page wasn't reliable because of the link. Of course this could be more to do with my supervisors POV than the general case. Ultima 10:20, 14 November 2006 (UTC)
Well, I only said that would be interesting to keep the link about bionics, because both Professor Ingo Rechenberg and Professor Hans-Paul Schwefel (two ES pioneers, the third is Peter Bienert) consider that their brainchild (evolution strategy) was also inspired by the cybernetics Zeitgeist along the 1960s - and cybernetics has a connection with bionics. If you verify the books written by them, you would realize the connection between those subjects.
Of course, if I read some very specific Wikipedia article related to genetics, probably I would not find any link to Biology, but I would know that I am reading something which has to do with Biology. Maybe the same could be applied to ES, which is quit a bit unknown to the general audience (even inside the evolutionary computation field, ES is sometimes neglected). So, although I am reading a very specific wikipedia article, that is, evolution strategy, would be nice to know that ES is just a part of a major field.
I recognize that would be much more interesting linking to Evolutionary computation (or any other EC-related stuff) rather than to Bionics. However, this last link would be of historic interest for anyone who wants to know the past and origins of evolution strategies. :) Nosophorus 01:02, 16 November 2006 (UTC)
I agree with that, but it might be better to put the link in the article itself rather than "See also", maybe include a paragraph in the text about the early work and it's rooting in other Bionics research, as above. TBH it would be nice to have more information on this page anyway :). For me "See also" refers to something comparable or tightly related to the current article, not anything that's related, but that's just my opinion ;) If you look at this page this is far more what I would expect to see in the "See also" section http://mathworld.wolfram.com/EvolutionStrategies.html Also, the Japanese version of this article links to Evolutionary algorithm, Genetic Algorithm, Genetic Programming, Evolutionary programming they don't even have a Bionics page. Ultima 13:17, 16 November 2006 (UTC)
Maybe add a section about the history and major research periods? It's not only in the 1960's this was big. Ultima 13:26, 16 November 2006 (UTC)
History
Early Work
Professor Ingo Rechenberg and Professor Hans-Paul Schwefel (two ES pioneers working in the field of Bionics, the third is Peter Bienert) consider that their brainchild (evolution strategy) was also inspired by the cybernetics Zeitgeist along the 1960s.

[edit] Recombinative ES == Real Coded GA?

Could you say where you found that

Contemporary derivatives of evolution strategy often use a population of μ parents and also recombination as an additional operator (called (μ/ρ+,λ)-ES). This is believed to make them less prone to get stuck in local optima. Because of using recombination, these algorithms might also be classified as real-coded genetic algorithms.?


If you have some reference (paper) to justify that argument, please, show us. :)


The first point that should be noticed is related to the selection schemes applied in Genetic Algorithms (GAs) and those used in Evolution Strategies (ESs). In GAs, the selection scheme applied to select the individuals which will be evolved usually is fitness based (tournament, roulette, etc), but in ESs all individuals are evolved, even those that are not so fit.

I consider that argument (bold text) is incorrect because ES has a very particular kind of recombination operators, which even have convergence and progress rate analysis. However, inside GA field there are not recombination operators like those.

And more: In ES the evolutionary process happens in the order recombination-mutation-selection, while in GA it is selection-recombination-mutation. Traditionally, the recombination operator is the major one in GAs, while in ESs that role is played by the mutation operator.

An important point: In ES, the recombination operator can work with 1 \le \rho \le \mu. What means that from 1 up to μ individuals can be involved in crossover to produce just one offspring. In binary GA, there are generally 2 chromossomes that generate 2 offsprings. So, here we have another difference between GA and ES. Also, in ES all μ parent individuals can be mutated and recombinated, while in GAs some parents are not selected to be evolved by the evolution process, that is, by the evolutionary operators.

Another point that should be considered is the fact that Evolution Strategies use the self-adaptation mechanism to tune its parameters, while in Genetic Algorithms we do not have that kind of adaptation.

There were some critiques about using real-coded GA (big alphabets), because binary representation help[s] convince us of the greater information that becomes available using smaller alphabets (Goldberg, Real-coded genetic algorithms virtual alphabets and blocking, 1990, pp 4). Although also there are some critiques related to GAs, both real- and binary-coded ("Reevaluating Genetic Algorithm Performance under Coordinate Rotation of Benchmark Functions; A survey of some theoretical and practical aspects of genetic algorithms." BioSystems, 39(3):263-278, 1996, Elsevier Science.). But these latter ones are quit a bit "unpopular".

So, after all those points I listed and explained, I consider that would be sensible to delete that part of the text that says recombinative evolution strategies might be classified as real-coded genetic algorithms.

If someone has some critique about what I wrote, please, tell me. :)


References

Nosophorus 18:36, 11 November 2006 (UTC)

Nosophorus 19:43, 2 January 2007 (UTC)

Nosophorus 04:14, 3 January 2007 (UTC)