An Improved CMA-ES for Solving Large Scale Optimization Problem

In solving large scale optimization problems, CMA-ES has the disadvantages of high complexity and premature stagnation. To solve this problem, this paper proposes an improved CMA-ES, called GI-ES, for large-scale optimization problems. GI-ES uses all the historical information of the previous generation of individuals to evaluate the parameters of the distribution of the next generation. These estimates can be considered as approximate gradient information, which complete covariance information is not required. Thus GI-ES is friendly to large scale optimization problems. Comparative experiments have been done on state-of-the-art algorithms. The results proved the effectiveness and efficiency of GI-ES for large scale optimization problems.