Distributed Cooperative Co-Evolution with Adaptive Computing Resource Allocation for Large Scale Optimization
Ya-Hui Jia,Wei-Neng Chen,Tianlong Gu,Huaxiang Zhang,Hua-Qiang Yuan,Sam Kwong,Jun Zhang
DOI: https://doi.org/10.1109/tevc.2018.2817889
IF: 16.497
2019-01-01
IEEE Transactions on Evolutionary Computation
Abstract:Through introducing the divide-and-conquer strategy, cooperative co-evolution (CC) has been successfully employed by many evolutionary algorithms (EAs) to solve large-scale optimization problems. In practice, it is common that different subcomponents of a large-scale problem have imbalanced contributions to the global fitness. Thus, how to utilize such imbalance and concentrate efforts on optimizing important subcomponents becomes an important issue for improving performance of cooperative co-EA, especially in distributed computing environment. In this paper, we propose a two-layer distributed CC (dCC) architecture with adaptive computing resource allocation for large-scale optimization. The first layer is the dCC model which takes charge of calculating the importance of subcomponents and accordingly allocating resources. An effective allocating algorithm is designed which can adaptively allocate computing resources based on a periodic contribution calculating method. The second layer is the pool model which takes charge of making fully utilization of imbalanced resource allocation. Within this layer, two different conformance policies are designed to help optimizers use the assigned computing resources efficiently. Empirical studies show that the two conformance policies and the computing resource allocation algorithm are effective, and the proposed distributed architecture possesses high scalability and efficiency.