Offspring are new candidate solutions created from existing individuals in an evolutionary algorithm. Like biological offspring, they are not necessarily copies of their parents: they can inherit a mix of traits and acquire small random changes. This gives the algorithm new possibilities to test. section 4.1 section 4.11
A genetic algorithm creates offspring as part of a repeating cycle:
- Score the current candidates with a fitness function, then select parents. section 4.3
- Use crossover to mix parts of the parents’ chromosomes; use mutation to change offspring slightly. section 4.11
- Choose which existing individuals and offspring enter the next generation. section 4.3
Mixing two parents
In two-point crossover, choose two positions in the parents’ chromosomes. An offspring takes the first and last parts from one parent and the middle part from the other—like splicing arrays. Another offspring can take the inverse mix. Each result must still represent a valid solution. section 4.13 section 4.20
Figure 4.19 Reproduce offspring.
Shows how reproduction combines parents and then introduces variation in their offspring. section 4.11
How many offspring to create is a design choice. More can increase diversity, but in a fixed-size population they also require more individuals to be removed, potentially including good solutions. A steady-state model might keep 80 of 100 individuals and add 20 offspring; a generational model could replace all 100. Offspring are possibilities to evaluate, not guaranteed improvements. section 4.20 section 4.9
