ANNEvolve :: Evolution of Artificial Neural Networks

There is a population of similar, but not identical, things. In the case of ANNEvolve software, these things are Artificial Neural Networks (ANNs).
Each one has a linear array of numbers that specify how it differs from the other members of the population. This array may be called a chromosome.
There is a test that can be applied to every member of the population, and which yields a fitness value. This fitness value is a measure of how each individual compares with the others in achieving some goal.
The entire population is tested, and then a significant fraction of them are chosen to be eliminated. The decision in each case is based on the results of the fitness test. It might just consist of eliminating the lower half of the population. More commonly there is random elimination with the less fit individuals having a higher probability of being eliminated.
There is mating, which consists of choosing pairs of surviving individuals to be parents. Usually, the more fit individuals have a higher probability of being chosen. Individuals may be parents more than once, and with different partners.
There is reproduction, which consists of creating new individuals to replace those that have been eliminated. These "offspring" are formed by combining portions of the chromosomes of both parents. Two new chromosomes are created which are different from both of the parents, but contain long segments of chromosome that are identical to homologous segments in one parent or the other. The total population is usually kept constant.
There is mutation, which consists of making some random changes to some parts of the chromosome. If the chromosome is binary, some bits are chosen at random to be toggled. If it's a string of floats, then some or all of them are changed by random amounts.
Steps 4 through 7 are repeated indefinitely, perhaps thousands of times. The result is usually a dramatic increase in the fitness of the population, and of the most fit member of the population. This most fit member will be usually be the desired solution to the problem at hand. Each repetition of these steps is called a generation.

Meet Annevolve's skydiving, mouseball collecting Unix Admin

Author: Eric Anderson
Posted: 11/14/2003

Species Learning and a Hypothesis About Brain Learning

Author: Mitchell Timin
Posted: 10/02/2003

When doing GA, expect a very large variance in the time required to accomplish a certain amount of evolution.

Author: Kent Pault Dolan
Posted: 09/09/2003

An Aspect of Natural Evolution

Author: Mitchell Timin
Posted: 08/31/2003

Genuine Artificial Intelligence :)

Author: Mitchell Timin
Posted: 08/27/2003