Exploiting disruption aversion to control code bloat

The authors employ multiple crossovers as a novel natural extension to crossovers as a mixing operator. They use this as a framework to explore the ideas of code growth. Empirical support is given for popular theories for mechanisms of code growth. Three specific algorithms for multiple crossovers are compared with classic methods for performance in terms of fitness and genome size. The details of the performance of these algorithms is examined in detail for both practical value and theoretical implications. The authors conclude that multiple crossovers is a practical scheme for containing code growth without a significant loss of fitness. Categories and Subject Descriptors Genetic Programming [] General Terms Algorithms, Experimentation, Performance, Theory Keywords code bloat, code growth, effective fitness

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Jason Stevens, Robert B. Heckendorn, Terence Soule
• Keywords: Code Growth, GECCO 2005, Keywords Code Bloat, Multiple Crossovers