作者: Riccardo Poli , William B. Langdon , James A. Foster , Tery Soule
DOI:
关键词:
摘要: The phenomenon of growth in program size genetic programming populations has been widely reported. In a variety experiments and static analysis we test the standard protective code explanation find it to be incomplete. We suggest bloat is primarily due distribution fitness space possible programs because this, absence bias, general inherent any search technique using variable length representation. investigate landscape produced by tree-based operators when acting upon points space. show common exponential shape underlying Nevertheless demonstrate new with considerably reduced bloating characteristics. also describe mechanisms whereby arises relate these back Finally our simple random walk entropy increasing model able predict evolved programs.