High-Dimensional Optimization using Diagonal and Low-Rank Evolution Strategy

In recent years, with the rise of artificial intelligence, large-scale optimization problems have become more and more popular. Evolution strategies (ESs) are a family of successful randomized algorithms for black-box optimization. Recent years have seen an increasing interest of designing efficient algorithms for high-dimensional problems. We propose a novel variant of the covariance matrix adaptation evolution strategy (DRx-ES) using a covariance matrix parameterized with a smaller number of parameters for solving high-dimensional problems. We demonstrate the performance of our proposed algorithm on bbob-largescale suite implemented in COCO and and compare it to different algorithms. We further evaluate the algorithm on the CEC'2010 large scale global optimization benchmarks with different degrees of non-separability, and show that DRx-ES outperforms all the compared algorithms.