Modelling Bounded Rationality using Evolutionary Techniques - Bruce Edmonds

8 Discussion

Such modelling using evolutionary techniques, where there is an explicit one-one correspondence between items modelled and the genes in the population typically deal with very small populations (in evolutionary terms). In the example above we had populations of mental models as small as 10. Most of the models of abstract evolutionary algorithms deal only with large populations (many assume an infinite population for formal purposes). The behaviour of small populations may be pathological from the point of view of an efficient search mechanism, but here we have different goals in using evolutionary algorithms. It is precisely the pathological aspects of the process that capture the qualitative behaviour observed: sharp path-dependence, lock-in, exploitative search, a large spread of behaviours between different populations and limited overall optimization.

Also it is not always the case that the usual genetic operators are very efficient in such small populations. It is known that selective breeding can work well with small populations [8]. In addition (in the example above) we found that a traditional GP mixture of tree-crossover and propagation did substantially worse than that of combining together old models, generating small new random ones and propagation. This is important as the mechanism chosen has to be credible for realistically small populations of mental models. Much work needs to be done to understand the evolutionary dynamics of small populations.

Modelling Bounded Rationality using Evolutionary Techniques - Bruce Edmonds - 09 JUN 97

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

Generated with CERN WebMaker