Interval-fitness interactive genetic algorithms with varying population size based on semi-supervised learning
基于半监督学习的变种群规模区间适应值交互式遗传算法

In order to alleviate user fatigue and improve the performances of interactive genetic algorithms (IGAs) in exploration, we present the interval-fitness interactive genetic algorithms with varying population size based on a cotraining semi-supervised learning(CSSL). According to the clustering results of a large population, we develop the strategy for selecting unlabeled samples and labeled samples. Based on the approximation precision of two co-training learners, an efficient strategy for selecting high reliable unlabeled samples for labeling is given. Then, the CSSL mechanism is employed to train two radial basis function(RBF) neural networks in order to establish the surrogate model with high precision and good generalization ability. In the subsequent evolution, the surrogate model is used to estimate the fitness of an individual; in turn, the surrogate model is updated based on its estimation error. The proposed algorithm is analyzed and applied to a fashion evolutionary design system. The experimental results show its efficacy.