These approaches can be viewed as performing a randomized
**beam search**, ie. a search that always keeps its best k hypotheses.

- Initialize the population to contain s individuals
- while (stopping condition not satisfied)
- Select (1-r)s individuals at random from the current generation according to fitness
- Select another rs/2 pairs of individuals at random from the current generation. Each pair participates in crossover, producing two offspring.
- The next generation consists of the (1-r)s individuals selected in the first step plus the rs offspring from the second step
- Mutate the next generation at rate m

Machine Learning

Tom M. Mitchell

McGraw-Hill, 1997