A Self-adaptive Multi-task Differential Evolution Algorithm

Abstract

This paper proposes a new self-adaptive scheme and differential evolution based evolutionary multi-task optimization algorithm to address multiple different optimization problems or tasks simultaneously. The proposed algorithm assigns a specific population and a transfer rate for each task and uses the differential evolution strategies to update each population. Compared with traditional evolutionary multi-task optimization algorithms that adopt a fixed transfer rate, the proposed algorithm uses a self-adaptive scheme to dynamically adjust transfer rate, which reduces the harm of negative transfer on the evolutionary direction of the population. The population is simultaneously driven by the information from the intra-task and other tasks. Based on the performance of the two evolutionary strategies, the population adaptively adjusts the transfer rate to complete the high-quality knowledge transfer process. The experiment is conducted on the single-objective multi-task test suite. The results show that the proposed algorithm can find more accurate solutions with a faster convergence rate in comparison with several state-of-the-art evolutionary multi-task optimization algorithms.