Evolving exact integer algorithms with Genetic Programming

摘要: The synthesis of exact integer algorithms is a hard task for Genetic Programming (GP), as it exhibits epistasis and deceptiveness. Most existing studies in this domain only target few simple problems or test small set different representations. In paper, we present the (to best our knowledge) largest study on to date. We first propose novel benchmark suite 20 non-trivial with variety features. then two approaches reduce impact negative features: (a) new nested form Transactional Memory (TM) epistatic effects by allowing instructions program code be permutated less behavior (b) recently published Frequency Fitness Assignment method (FFA) chance premature convergence deceptive problems. full-factorial experiment six loop instructions, TM, FFA, find that GP able solve all problems, although not them high success rate. Several interesting are discovered. FFA has tremendous positive while TM turns out useful.