.png){kind=icon}
.png){kind=icon}
Return
Multifactorial differential evolution enhanced by adaptive Gaussian-mixture-model-based knowledge transfer
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2025.102194.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Humans rarely tackle every task from scratch. Consequently, common knowledge gained from different (but potentially related) optimization tasks can aid in solving one another. Along this direction, evolutionary multitasking optimization (EMTO) has been proposed to solve multiple optimization tasks simultaneously, with the goal of improving performance on each task compared to solving them independently through knowledge transfer. Although the multifactorial evolutionary algorithm (MFEA), a representative EMTO algorithm, has achieved some success, it often suffers from slow convergence and negative knowledge transfer, especially when the similarity among tasks is low. To address this problem, this study presents a novel multifactorial differential evolution equipped with adaptive model-based knowledge transfers (named MFDE-AMKT for short). A Gaussian mixture model (GMM) is employed, in which a Gaussian component is used to capture the subpopulation distribution of each task. Unlike existing studies, the mixture weight and mean vector of each subpopulation distribution in the GMM are adaptively adjusted to fit the current evolutionary trend. Specifically, the mixture weights of the GMM are determined based on the overlap degree of the probability densities on each dimension, enabling a fine-grained measurement. Additionally, the mean vector of each subpopulation distribution is adaptively adjusted to explore more promising areas. Experimental studies conducted on both single-objective and multi-objective multi-task test suites have demonstrated the effectiveness and efficiency of the proposed MFDE-AMKT in comparison with several state-of-the-art evolutionary algorithms.
Journal
IF:
8.5
Papers:
2.1K
Citations:
1.0W
Organization
View more organizations