Title: Optimisation of the high-order problems in evolutionary algorithms: an application of transfer learning
Authors: Guo-Sheng Hao; Gai-Ge Wang; Zhao-Jun Zhang; De-Xuan Zou
Addresses: School of Computer Science and Technology, Jiangsu Normal University, Xuzhou, Jiangsu, China ' School of Computer Science and Technology, Jiangsu Normal University, Xuzhou, Jiangsu, China ' School of Electrical Engineering and Automation, Jiangsu Normal University, Xuzhou, Jiangsu, China ' School of Electrical Engineering and Automation, Jiangsu Normal University, Xuzhou, Jiangsu, China
Abstract: Evolutionary Algorithms (EAs) have been applied to many optimisation problems, among which those with high order are difficult for EAs. The higher the order, the steeper the curve around the optimum is, therefore the more difficult it is. This paper introduces Transfer Learning (TL) aided EAs to conquer the high-order problems more efficiently and effectively by optimum transfer from the low-order problem (as source domain) to high-order problem (as the target domain). The experiments validated this method by comparison of the average number of the convergence generation and an impressive feature was observed: this method is robust against the difficulties of the problems. This method is not only significant for high-order problems, but also useful for other difficult problems by borrowing optimum from other feature-similar easy problems.
Keywords: evolutionary algorithm; optimum; difficult problem; transfer learning; high order.
International Journal of Wireless and Mobile Computing, 2018 Vol.14 No.1, pp.56 - 63
Available online: 13 Feb 2018 *Full-text access for editors Access for subscribers Free access Comment on this article