Genetic Programming Full Seminar Topic Free Download-Abstract

In theory, the task of mutation in GP is the same as in all other EC branches, creating a new individual from an old one through some small random variation. The most common implementation works by replacing the subtree starting at a randomly selected node by a randomly generated tree. The newly created tree is usually generated the same way as in the initial population, see Section 8. 

Note, that the size (tree depth) of the child can exceed that of the parent tree. Figure 5 illustrates how the parse tree belonging to the formula 1 (left) is mutated into a parse tree standing for 2 _ ¡ + ((x + 3) _ y)

Mutation in GP has two parameters:

The probability of choosing mutation at the junction with recombination, The probability of choosing an internal point within the parent as the root of the subtree to be replaced.

It is remarkable that Koza's classic book on GP from 1992, cf. [5], advises to set the mutation rate at 0, i.e., it suggests that GP works without mutation. More recently Banzhaf et al. suggested5% [2]. In giving mutation such a limited role, GP differs from other EA streams. The reason for this is the generally shared view that crossover has a large shuffling effect, acting in some sense as a macromutation operator [1]. The current GP practice uses low, but positive, mutation frequencies, even though some studies indicate that the common wisdom favoring an (almost) pure crossover approach might be misleading [9].

GENETIC PROGRAMMING FULL SEMINAR DOWNLOAD HERE