Selection

Selection is the process in an evolutionary algorithm that chooses candidate solutions from a population to produce offspring. It uses fitness to influence which candidates reproduce, balancing preference for strong solutions against continued population diversity.

Fitness function: A function that assigns a score to a candidate solution according to how well it satisfies an optimization problem. Its scores provide the basis for preferring some candidates over others.
Evolutionary algorithm: A stochastic optimization method that evolves a population of candidate solutions through variation and selection. Selection is one of the defining operations in this broader family of methods.
Random search: An optimization method that samples candidate solutions without using their fitness to guide the search. It removes fitness-based preference, unlike selection.
Genetic algorithm: An evolutionary optimization algorithm that represents solutions as encoded chromosomes and applies selection, crossover, and mutation. Parent choice determines which encoded traits enter its offspring.
Premature convergence: An optimization failure in which a population loses diversity and settles around a suboptimal solution too early. Excessive preference for current winners can accelerate this loss of alternatives.
Roulette wheel selection: A selection method that chooses candidates with probabilities proportional to their nonnegative fitness scores. It converts relative fitness directly into chances of reproduction.
Population (evolutionary computation): A collection of candidate solutions maintained during an evolutionary algorithm's search. Selection chooses reproductive candidates from this changing collection.
Truncation selection: A selection method that allows only candidates above a fitness threshold or within a top-ranked fraction to reproduce. Its hard cutoff contrasts with schemes that give weaker candidates some chance.
Differential evolution: A population-based optimization algorithm that generates trial vectors from scaled differences among candidate vectors. Its replacement decisions compare trial candidates with existing population members.
Exploration and exploitation: The tradeoff between searching unfamiliar possibilities and refining known promising ones. Selection pressure shifts evolutionary search toward exploitation or continued exploration.

Linked from 2 pages