Genetic algorithms: evolving behaviour

≡放荡痞女 提交于 2019-12-03 20:39:01

As Gordon Murray Dent said, when evolutionary programming is used to create functions, it's generally referred to as "genetic programming" while "genetic algorithms" typically refers (somewhat oddly) to function parameter optimization.

The classic texts on genetic programming are Koza's books, which are still VERY relevant especially if you happen to be a LISP developer. His Website http://www.genetic-programming.com/ should give you plentiful leads for other resources.

Check out genetic programming.

Good thread on Google Answers about some algorithms Google answers

Apache as a project named Mahout it's having a lot of good documentation about algorithms and technics they used.

"Machine learning algorithms" is a good keyword to search in google.

You can stick with optimization of pure parameters, yet obtain behaviors... Just evolve neural networks. For instance, a multi-layer Perceptron is able to approximate all functions. So evolve its parameters, such as the function it computes fullfill your objective. And tada, if you use a decent algorithm and don't have more than, say, 100 parameters, you might get something nice. Note that a neural network have real parameters, so use an evolutionary algorithm designed for real parameters. I suggest CMA-ES, not much tweaking necessary, the current benchmark winner & used in the real world.

I used the GP field guide for my thesis. It is reasonably useful when you want to learn the basics of genetic programming. It contains contributions from John Koza. The book is freely available and it has a lot of references to papers and other books if you need some specific in-depth information. It is a very good reference.

As for "classical" genetic algorithms I got recommended the book Genetic Algorithms + Data Structures = Evolution Programs, I haven't read it though.

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!