Scalable model based evolutionary multi-objective optimization

This talk presents some recent advances in model-based evolutionary multi-objective optimization. We first present a regularity based estimation of distribution algorithm that uses a probabilistic model containing a principal curve and a local Gaussian model. We show that the proposed algorithm is able to work efficiently for large dimensional optimization problems with a small population size. Then, a new evolutionary multi-objective optimization algorithm by using inversed modeling is discussed. By means of constructing inverse models, we are able to sample preferred solutions in the objective space, which is proved to be helpful in maintaining the diversity and generating additional solutions at a low computational cost.