Surrogate Models Based on Individual's Interval Fitness in Interactive Genetic Algorithms

We presented a novel interactive geneticalgorithm with surrogate models based on an individual'sinterval fitness in this paper. In this algorithm, the surrogate models are generated based on an individual's intervalfitness. The adopted surrogate models are two radial basis function networks to model the upper limit and thelower limit of an individual's fitness, respectively. Having been trained using the gradient descent approach withdata obtained in the process of the evolutions, these modelsare then applied to estimate all individuals' fitness in thesubsequent evolutions. The surrogate models continuouslyupdate during the evolutions in order to improve their precision. We quantitatively analyzed the performance of thealgorithm in alleviating user fatigue and increasing the opportunity to look for the optimal solutions. In addition, wealso applied the algorithm to a fashion evolutionary designsystem. The results show that the proposed algorithm isadvantageous.