A Surrogate-Assisted Cooperative Co-evolutionary Algorithm Using Recursive Differential Grouping as Decomposition Strategy
Julien Blanchard, Charlotte Beauthier, Timotéo Carletti · 2019
Cooperative co-evolutionary algorithms, especially those able to uncover interaction structure between variables, have a great potential in optimizing large-scale problems. Nevertheless, they are expensive in terms of number of function evaluations and this issue can be quite problematic when dealing with computationally expensive optimization problems. An effective approach to deal with such problems lies in the exploitation of surrogate models. The latter ones work as cheap-to-evaluate alternatives to the expensive function reducing the computational cost, while still providing improved designs. This process, called surrogate-assisted optimization, is very effective on small-dimensional problems but is not suitable to solve large-scale problems due to the curse of dimensionality. In this paper, a new algorithm, taking benefit from cooperative coevolution and surrogate models, is introduced to efficiently solve high-dimensional, expensive and black-box problems. The proposed algorithm uses recursive differential grouping to perform an accurate problem decomposition. Experimental results are provided on a set of 1000-dimensional problems and show promising results.