Jump to content

Talk:Genetic algorithm

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia
This is an old revision of this page, as edited by Pierre-delmoral (talk | contribs) at 20:33, 14 September 2015 (→Removed content about genetic algorithms for Monte Carlo integration). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

The related algorithms section was once a huge, rambling list of unsorted, often obscure methods. I have attempted to tidy up the section as best as I can by using sub-headings to put each method into context. I have deleted nothing so far, but the section needs a good purge. Problems include:

  • Too many algorithms listed that are very weakly related to GA
  • Too many obscure algorithms - it seems that some researchers are perhaps self-promoting their work? Such methods should be deleted until the methods have wider acceptance in the community (and the linked wiki pages are updated consequently)
  • I think it's okay to have a short description by each algorithm in this section, but nothing more than a short sentence. More information on each algorithm should be obtained on the algorithm's main page.

Overall this section should really be no more than about 15 lines. Thoughts? Jr271 (talk) 19:10, 27 March 2011 (UTC)[reply]

Another example to add?

An evolving mechanical arm. It uses genetic algorithms to train a neural network. http://www.e-nuts.net/en/genetic-algorithms —Preceding unsigned comment added by 78.26.90.227 (talk) 13:54, 11 May 2011 (UTC)[reply]

Added to external links. 82.81.159.224 (talk) 06:46, 1 July 2013 (UTC)[reply]

An invitation to experiment

Another editor and I have reverted an insertion that includes an invitation to experiment.[1] The URL in the first link leads to directory where one should presumably download a demo to try out some poor computational bounds. The references are primary, do not seem to be on point, and seem to be more about promoting a particular author.

Please get a consensus before reinserting this material. Glrx (talk) 14:41, 20 June 2013 (UTC)[reply]

Hi there, not sure how an ordinary reader reader will know or care about this BRD thing? I'd just want a quick 'pedia reference. BTW, the URL appears to lead to an online experiment, not to a site to download a demo. 2401:7400:E800:C601:A5F9:65C3:533B:49F0 (talk) 18:04, 20 June 2013 (UTC)[reply]
Keep it out. An easily accessible, interactive demonstration might be a useful external link, but this doesn't appear to be one, nor should it be added in the body of the article as it was. --Ronz (talk) 18:48, 20 June 2013 (UTC)[reply]

I would go for the new version, as it gives useful references. These editors who reverted this new version seem not to have a real interest in Genetic Algorithms or know this stuff in depth. Sometimes I wonder what the best way would be for people to work seamlessly on useful references or information to benefit the Wikipedia reader. — Preceding unsigned comment added by 110.74.220.41 (talk) 14:33, 25 June 2013 (UTC)[reply]

To readers who happen to land on this consensus page

What is this consensus about? It is about a proposed addition for an updated Wikipedia of 'Genetic Algorithm', which tries to address the [citation needed] flag by providing a Java Applet online for the reader to experiment the upper and/or lower bounds for the mutation/combination and/or crossover parameters, although these parameter rates depend on whether or not to utilise direct inheritance. The website does not provide experiments on direct heredity, but a reference on it is made available in this updated version.

If you think that the new version offers a quick or useful reference, especially in the age of electronic encyclopaedia, or otherwise, do feel free to make your views known here (and then click on the 'save page' button below). 2401:7400:E800:7CCD:8538:484D:FEE0:E308 (talk) 08:24, 24 June 2013 (UTC)[reply]

Yes, the user can select various crossover and mutation rates before start to experiment easily on the effects and bounds of these parameters, as well as on the effect of an 'elite', by following the 'Background evolve' after 'Start'. — Preceding unsigned comment added by 175.156.139.134 (talk) 08:38, 24 June 2013 (UTC)[reply]
Why good tutorials are now missing from the current page? Was it because of the discussions here? I think the EA Demo tutorial and the new references here give details that fill the gap in the present version. Who are allowed to restore the previous page? Can we have the better page which is discussed here, please.

GA and natural selection

An anonymous user (IP 24.62.24.89) apparently dislikes the analogy between natural selection and genetic algorithms, and has been adding gratuitous comments at the end of the first sentence. I have already reverted one only to see a similar comment added back. I do not want to engage in a "revert war", so I'm just pointing it out as vandalism.

Piotr Gasiorowski (talk) 16:55, 15 January 2014 (UTC)[reply]

Thanks for monitoring the page and reverting. I will watch list it as well. --Mark viking (talk) 17:36, 15 January 2014 (UTC)[reply]

Intro

It seems odd to me that GA should only be inspired by "natural selection" and other evolutionary algorithms by "evolution". Crossover, reproduction, mutation are distinct features of GAs as well. --Robin to Roxel (talk) 16:39, 18 August 2015 (UTC)[reply]

Removed content about genetic algorithms for Monte Carlo integration

This version of the article contained text about Monte Carlo methods and signal processing, that were removed with this edit. I link them here per WP:PRESERVE. Diego (talk) 21:50, 23 August 2015 (UTC)[reply]

I completely agree that this unbalanced the article. Stuartyeates (talk) 09:42, 7 September 2015 (UTC)[reply]
From the probabilistic and statistical point of view, Genetic Algorithms with mutation and selection transitions can be interpreted as a natural acceptance rejection simulation technique equipped with a interacting recycling mechanism. Introduced in the 1950s these genetic type evolutionary Monte Carlo methods are used to sample complex and high dimensional probability distributions. When the number of individuals (and the computational power) tends to infinity, it can be proved that the occupation measures of the individuals converge to a Feynman-Kac measure on path space. These distributions arise in Bayesian inference, nonlinear filtering, rare event simulations, as well as in molecular chemistry, and stochastic optimization. In contrast to heuristic like genetic algorithms discussed in the literature on genetic algorithms, the genetic type Monte Carlo methods discussed in this article are mathematically well founded, and they allow to solve complex Monte Carlo integration problems. — Preceding unsigned comment added by 14.200.118.120 (talk) 00:46, 14 September 2015 (UTC)[reply]
This version of the article contained essential informations on the use of genetic algorithms for solving Monte Carlo integration problems arising in physics, chemistry, risk analysis, and signal processing. This article not only emphasize an avenue of new application domains of evolutionary computation, it also provides rigorous mathematical foundations of genetic algorithms with selection and mutation transitions as the size of the population tends to infinity. — Preceding unsigned comment added by Pierre-delmoral (talk • contribs) 04:18, 14 September 2015 (UTC)[reply]
Pierre. Thank you for contributing here, but please bear in mind that we aim for our articles to be accessible to lay-people rather than being rigorous academic sources. The content you added is impenetrable to the average reader who comes here wanting to know what a genetic algorithm is. Additionally, per WP:SELFCITE it is recommended that editors don't cite their own research as this creates problems, particularly in relation to our policy on no original research. If you wish to improve the article, please use sources such as text books or reviews. Thanks SmartSE (talk) 09:39, 14 September 2015 (UTC)[reply]


Smartse. The removed article doesn't describe any type of mysterious evolutionary algorithm nor any heuristic type impenetrable genetic algorithm. We have been working on the first rigorous mathematical foundations of genetic algorithms for Monte Carlo integration and their refined analysis since more than 20 years. So we were obliged to cite these pioneering studies, we also gave free accessible reviews, precise and verifiable links to related subjects, complementary and original works by other authors, as well as links to equivalent algorithms currently used in other scientific disciplines. The removed article also provides a more detailed history on the use of genetic algorithms in computational sciences. Of course, the reader who want to know about rigorous mathematical foundations need to have some knowledge on Monte Carlo integration and the law of large numbers. A reader with some basic background in these two subjects will understand without any difficulties the mathematical aspects of genetic algorithms when the size of the population tends to infinity. I thought an article on genetic algorithms should at least explain what happens when the computational power and the size of the population tends to infinity. I didn't knew it was preferable to have only a text describing all the heuristic cooking rules that can be used in practice.
Retrieved from "https://en.wikipedia.org/w/index.php?title=Talk:Genetic_algorithm&oldid=681042639"