Sexual Selection in Genetic Programming: Advertising Good Genes
研究了基于性选择的遗传编程方法，在符号回归任务中测试了两种装饰染色体变体，发现该方法在多数数据集上优于标准锦标赛选择和PIMP方法，并提高了种群多样性。
In sexually reproductive species, the quality of a mate is as important as survival itself, making sexual selection crucial for evolution, promoting phenotypic and behavioural diversity, either by biases towards traits or by signalling fitness advantages. Recently, Desire-Driven Selection has been proposed, modelled after mate choice, where individuals hold preferences and population-dependent ornaments that signal local fitness quality. In this work, we expand the analysis of the method by experimenting with a wider range of real-world datasets for symbolic regression, measuring performance, diversity levels, and running time. We test two variations of the ornament chromosome, using the median as a threshold for trait activation, and introduce a simplified version using cumulative error. Results show a clear improvement over the standard tournament selection, outperforming it in eight out of ten datasets, reaching statistical significance in seven instances, while increasing the overall population diversity by 26.1% on average. The approach also shows fitness improvements over PIMP, a mate choice-based approach using ideal mate representations, in nine datasets, achieving statistical significance in seven. In addition, the method consistently shows competitive results when compared against a state-of-the-art Lexicase selection, finding the best overall solutions, while also being faster to compute by 13% on average.