| Breeder genetic algorithm |
Basic evolutionary algorithm using truncation selection |
[74], [148]
|
| Phenotypic niching with fitness sharing |
The reproductive opportunities of individuals are shared amongst members of a niche. A niche is defined by a neighbourhood in phenotype-space, i.e. as a vector of attributes or traits. The scheme seeks to preserve diversity. |
[130], [149]
|
| Deterministic crowding |
Crowding is a reproduction scheme in which individuals are forced to replace individuals in the population that are most like them. In deterministic crowding this is achieved without inspection of the genotype; offspring merely replace their parents (depending on the relative fitness of the parents and offspring). Preserves diversity. |
[150], [151]
|
| Local selection; local breeding; local mating; spatially structured populations |
The population is given some spatial structure (usually independently of fitness), and mating is allowed to occur only between neighbours in this structure. Similarly, offspring replace low-fitness individual(s) within their own neighbourhood. Preserves diversity. |
[152]
|
| Island model GAs (Alba and Tomassini) |
Several populations evolve on separate islands using locally panmictic mating. There is limited but occasional migration from one island to the other. Preserves diversity. |
[54], [153]–[155]
|
| Landscape state machine tuning of directed evolution (LSM-DE) |
In this technique, choices for mutation rate, population size, selection pressure and other evolutionary parameters are based on some prior sampling – and subsequent modelling – of the fitness landscape (or one believed to have similar topological features). Tunes the search algorithm specifically to the problem. |
[90], [156], [157]
|
| Fitness uniform selection scheme (FUSS) |
FUSS is a selection scheme that preserves phenotypic diversity. |
[158], [159]
|
| Hybrid local search or memetic algorithms |
Evolution scheme in which selected individuals (usually fitter ones) in each generation are improved by performing a fitness-directed walk on the landscape. Improves exploitation of fit individuals, driving them towards optima. May not be feasible for some types of Directed Evolution experiment. These may not perform well when large populations but small numbers of generations are available (since the adaptive walks necessarily take the equivalent of several breeding generations to complete.) |
[160], [161]
|
| Statistical Racing |
Several evolutions, each with different parameters controlling selection pressure, mutation rates, and so on, are run simultaneously. At intervals, any evolution that is performing statistically significantly worse is dropped and its resources are allocated equally to the others. Expensive but effective. |
[162], [163]
|
| Self-tuning evolutionary algorithms |
Mutation rates, selection pressure, rates of recombination are controlled during evolution. These may be changed deterministically according to a schedule; changed according to some rules based on the progress being made; or actually evolved by making the parameters themselves subject to selection and variation. |
[60]
|
