Reinforcement learning-based adaptive operator selection

Durgut, Rafet; Aydin, Mehmet Emin

Abstract

Metaheuristic and swarm intelligence approaches require devising optimisation algorithms with operators to let produce neighbouring solutions to conduct a move. The efficiency of algorithms using single operator remains recessive in comparison with those with multiple operators. However, use of multiple operators require a selection mechanism, which may not be always as productive as expected; therefore an adaptive selection scheme is always needed. In this study, an experience-based, reinforcement learning algorithm has been used to build an adaptive selection scheme implemented to work with a binary artificial bee colony algorithm in which the selection mechanism learns when and subject to which circumstances an operator can help produce better and worse neighbours. The implementations have been tested with commonly used benchmarks of uncapacitated facility location problem. The results demonstrates that the selection scheme developed based on reinforcement learning, which can also be named as smart selection scheme, performs much better that state-of-art adaptive selection schemes.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 4th International Conference on Optimization and Learning
• Start Date: Jun 21, 2021
• End Date: Jul 23, 2021
• Acceptance Date: Mar 30, 2021
• Online Publication Date: Aug 17, 2021
• Publication Date: 2021
• Deposit Date: Aug 18, 2021
• Publicly Available Date: Aug 18, 2021
• Publisher: Springer Verlag (Germany)
• Volume: 1443
• Pages: 29-41
• Series Title: Communications in Computer and Information Science
• Series Number: 1443
• Series ISSN: 1865-0929
• Chapter Number: 3
• ISBN: 9783030856717
• DOI: https://doi.org/10.1007/978-3-030-85672-4_3
• Public URL: https://uwe-repository.worktribe.com/output/7638600