Any algorithm that solves a problem by picking a bunch of solutions at random, calculating the fitness of each solution, and then choosing a new set of solutions that have something in common with the better solutions of the previous set. This process is then repeated indefinately, and the average fitness of each new group of solutions will be generally higher than that of the previous group. Eventually, each new set of solutions will not vary that much from a certain solution, no matter how long the algorithm continues to run. It is then said to have converged on that solution. If the algorithm was well-suited to solving the problem, this will be the optimal solution. Otherwise, it will have reached an evolutionary dead end.

The process of genetic recombination in biological life forms is an example of an evolutionary algorithm, whereas genetic algorithms, evolutionary strategies, and classifier systems are all examples of artificial evolutionary algorithms.

Log in or register to write something here or to contact authors.