[Next] [Previous] [Top] [Contents]
Meta-Genetic Programming: Co-evolving the Operators of Variation - Bruce Edmonds
There are good reasons to suppose that there is no one technique, however clever, recursive, or self-organising will be optimal for all problem domains [11]. There always seems to be an implicit trade-off between the sophistication of a technique and its computational cost. Much of the power of GP comes from its low computational cost compared to its effectiveness, allowing large populations to successfully evolve solutions where more carefully directed algorithms have failed. Thus the conditions of application of MGP are very important - when it is useful to use and when not. I conjecture that techniques such as MGP might be helpful in finding answers to difficult problems but worse over simpler ones, but this (like the useful conditions of application of GP itself) is an open question.
A second advantage of GP is its robustness. MGP, through its nature, is likely to be far more brittle (as was illustrated by the various techniques needed to compensate for the inherent biases in the operator language). This is likely to be even more of a problem where more levels of populations acting on each other are involved or where populations act upon themselves.
Meta-Genetic Programming: Co-evolving the Operators of Variation - Bruce Edmonds - 28 JAN 98
[Next] [Previous] [Top] [Contents]
Generated with CERN WebMaker