Testing the Robustness of the Genetic Algorithm on the Floating Building Block Representation

Recent studies on a oating building block representation for the genetic algorithm (GA) suggest that there are many advantages to using the oating representation. This paper investigates the behavior of the GA on oating representation problems in response to three diierent types of pressures: (1) a reduction in the amount of genetic material available to the GA during the problem solving process, (2) functions which have negative-valued building blocks, and (3) random-izing non-coding segments. Results indicate that the GA's performance on oating representation problems is very robust. Signiicant reductions in genetic material (genome length) may be made with relatively small decrease in performance. The GA can eeectively solve problems with negative building blocks. Randomizing non-coding segments appears to improve rather than harm GA performance.