next up previous
Next: Acknowledgment Up: 3 Previous: Future Work



This chapter has tried to convey three aspects of a paradigm: what it is, how it works, and an intuition for why it works. The real point of the chapter is to convince other researchers that co-evolution and application of SMART operator programs can be made more effective than human-designed operators with minimal extra effort.

The experiments in section 3.5 demonstrate the ability of co-evolved SMART operators to outperform random recombination operators in a variety of domains. These experiments show absolute performance and percentage real improvement gains due to the SMART operators. In addition, the experiments show two main population attributes, one bad and one good, that were reduced and increased respectively over random operator performance, even though the operator fitness measure took neither attribute explicitly into account.

This chapter is not intended to be a showcase for impressive results. Rather, we hope to have conveyed a strong intuition about why there is much to gain and almost nothing to lose by incorporating SMART operators into the general framework of genetic programming. There are numerous possible additions to and adaptations of the SMART operator paradigm, many of which were mentioned in this chapter. We hope this chapter is the first word, not the last, a primer and inspiration, not a bible or an answer, on SMART operator program evolution.

Eric Teller
Tue Oct 29 17:04:55 EST 1996