27 April 2022 in IEEE Transactions on Evolutionary Computation
Decomposition plays a significant role in cooperative coevolution, which shows great potential in large-scale black-box optimization (LSBO). However, current learning-based decomposition algorithms require many fitness evaluations (FEs) to detect variable interdependencies and encounter the difficulty of threshold setting. To address these issues, this study proposes an efficient adaptive differential grouping (EADG) algorithm. Instead of homogeneously tackling different types of LSBO instances, EADG first identifies the instance type by detecting the interdependencies of a few pairs of variable subsets. Only if the instance is partially separable dose EADG further engage with it by converting its decomposition process into a search process in a binary tree. This facilitates the systematic reutilization of evaluated solutions so that half the interdependencies can be directly deduced without extra FEs. To promote the decomposition accuracy, EADG specially designs a normalized interdependency indicator that can adaptively generate a decomposition threshold according to its ordinal distribution. Theoretical analysis and experimental results show that EADG outperforms current popular decomposition algorithms. Further tests indicate that it can help cooperative coevolution achieve highly competitive optimization performance.
27 February 2020 in IEEE Access
Taking “divide-and-conquer” as a basic idea, cooperative coevolution (CC) has shown a promising prospect in large scale global optimization. However, its high requirement on the decomposition accuracy can hardly be satisfied in practice. Directing against this issue, this study proposes a bi-hierarchical cooperative coevolution (BHCC), which can tolerate a certain degree of decomposition error. Besides the cooperation among sub-problems as in the conventional CC, BHCC introduces a kind of cooperation between sub-problems and the overall problem. By systematically exploiting the excellent sub-solutions obtained during the sub-space optimization process, it initializes the population for the optimization process on the overall problem and thus can conduct search in promising regions of the whole solution space. The newly acquired complete solutions are in turn employed to update the context vector and the population of each sub-problem, where the context vector is used for sub-solution evaluation. Consequently, the search direction misdirected by an improper decomposition can be corrected to a great extent. To keep the balance between the two types of optimization processes, an adaptive triggering mechanism for the overall optimization process is specially designed for BHCC. Experimental results on two widely-used benchmark suites verify the effectiveness of the new strategies in BHCC and also indicate that BHCC is more robust than existing CCs and can achieve competitive performance compared with several state-of-the-art algorithms.
30 November 2022 in Information Sciences