Evolutionary systems & genetic algorithms

__FORCETOC__

__FORCETOC__

The following cycle of genetic competition lies at the core of evolutionary computing:

The following cycle of genetic competition lies at the core of evolutionary computing:

* A number of individuals (chromosomes or sets of genes) are in the population (gene pool) and each has a fitness score.

* A number of individuals (chromosomes or sets of genes) are in the population (gene pool) and each has a fitness score.

For bit-string genetic representations genetic variations are fairly straightforward.

For bit-string genetic representations genetic variations are fairly straightforward.

For engineering solutions with structured (non-bitstring) genes the distribution of mutations may be non-uniform, e.g. Gaussian (or Normal). Some studies have shown such techniques provide faster more accurate solutions.

For engineering solutions with structured (non-bitstring) genes the distribution of mutations may be non-uniform, e.g. Gaussian (or Normal). Some studies have shown such techniques provide faster more accurate solutions.

In addition the rate of mutation may be a genetic variable. This provides some relief in deciding what the mutation rate should be. The optimal mutation rate will then evolve. It's even possible that the mutation rate will self-adjust over time. Typically at the start the mutation rate should be higher (randomization) than it is later (when fine tuning). It's worth noting that some cases have been found in nature where the mutation rate increases in response to increased environmental pressure. This too could be simulated by calculating the mutation rate as a function of both genetics and environmental pressure.

In addition the rate of mutation may be a genetic variable. This provides some relief in deciding what the mutation rate should be. The optimal mutation rate will then evolve. It's even possible that the mutation rate will self-adjust over time. Typically at the start the mutation rate should be higher (randomization) than it is later (when fine tuning). It's worth noting that some cases have been found in nature where the mutation rate increases in response to increased environmental pressure. This too could be simulated by calculating the mutation rate as a function of both genetics and environmental pressure.

Note: In the first chart the tallies for Gray code is missing one line. "2 x 15" should follow "2 x 13".

Note: In the first chart the tallies for Gray code is missing one line. "2 x 15" should follow "2 x 13".

In both cases the mean degree of change is 3.75. In the case of binary codes there are as many transitions above the mean as below the mean. But in the case of gray codes 44 transitions are below the mean and 20 transitions are above the mean. Gray code provides a system where mutations result in a greater number of small changes, and a lesser number of large changes.

In both cases the mean degree of change is 3.75. In the case of binary codes there are as many transitions above the mean as below the mean. But in the case of gray codes 44 transitions are below the mean and 20 transitions are above the mean. Gray code provides a system where mutations result in a greater number of small changes, and a lesser number of large changes.

The goal of this classic optimization problem is to craft a path visiting each location once and only once in the shortest distance possible.  

The goal of this classic optimization problem is to craft a path visiting each location once and only once in the shortest distance possible.  

William Latham began working using a genetic approach even before using computers.

William Latham began working using a genetic approach even before using computers.

Working with sculptural forms, Latham's genes are variable length assemblies of modules which are connected with various operations for contour, scale, and twist.

Working with sculptural forms, Latham's genes are variable length assemblies of modules which are connected with various operations for contour, scale, and twist.

Each gene can be considered a parameter which contributes a dimension to a gene space. The more alike two individuals genes are the closer together they are in gene space.

Each gene can be considered a parameter which contributes a dimension to a gene space. The more alike two individuals genes are the closer together they are in gene space.

   

   

Using evolutionary methods, the artist applies an intuitive fitness function by manually evaluating phenotypes and assigning each a score. As a result the artist steers through the gene space towards an increasingly desirable result. As noted in the example above, attention must be paid to scaling so that the degree of variation is appropriate in each dimension. One approach is to normalize each parameter to values between 0 and 1, apply the variation method, and then re-map the parameter back into its native range.

Using evolutionary methods, the artist applies an intuitive fitness function by manually evaluating phenotypes and assigning each a score. As a result the artist steers through the gene space towards an increasingly desirable result. As noted in the example above, attention must be paid to scaling so that the degree of variation is appropriate in each dimension. One approach is to normalize each parameter to values between 0 and 1, apply the variation method, and then re-map the parameter back into its native range.

The artist typically uses a selective breeding tool to navigate through the gene space. Latham sculpts forms in the sense that a farmer sculpts increasingly healthy crops by developing hybrid seed.

The artist typically uses a selective breeding tool to navigate through the gene space. Latham sculpts forms in the sense that a farmer sculpts increasingly healthy crops by developing hybrid seed.

Here is an illustration of the tree structure technique with the expressions added.

Here is an illustration of the tree structure technique with the expressions added.

As noted by Sims:

As noted by Sims: