4 Algorithms

4.3. Evolutionary programming

GP search algorithms are very effective in finding an acceptable solutions in large search spaces - i.e. they perform an effective satisficing global search. However, they are less effective in searching locally to determine the best variation of a good solution once it has been found. That is, while GPs are excellent satisficers, they are not good optimizers.

For this reason we added a last stage to the search algorithm which is essentially a multiple stochastic hill-climbing algorithm using an evolutionary programming technique. This works by keeping the fittest half of the population in each generation as well as generating near mutations of each of them. Thus each generation the original and the mutation is compared and the better ones selected.

We applied this to the population that resulted from GP algorithm using a mutation algorithm adapted to real-valued parameters which mutates parameters according to a normal distribution, so as to favour near mutations above far ones. The standard deviation of this distribution was ramped downwards throughout this phase so that large mutations would be tried before progressively finer ones, in a similar manner to a simulated annealing algorithm.

Artificially Intelligent Specification and Analysis of Context-Dependent Attribute Preferences - 03 NOV 97

[Next] [Previous] [Up] [Top] [Contents]

Generated with CERN WebMaker