.png){kind=icon}
.png){kind=icon}
Return
EMT-DAKT: Evolutionary Multi-Objective Multi-Task Optimization Algorithm Using Deep Models With Adaptive Knowledge Transfer
.png){kind=icon}
PRE
AI
DOI：10.1109/TETCI.2026.3671021.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Evolutionary multi-objective multi-task optimization (MO-MTO) can optimize multiple tasks simultaneously by knowledge transfer (KT) between tasks. However, the existing MO-MTO algorithms still suffer from limitations in knowledge transfer flexibility and the quality of transferred knowledge, which ultimately leads to poor convergence performance. To address the above issues, this paper proposes a evolutionary multi-objective multi-task optimization algorithm using deep models with adaptive knowledge transfer, named EMT-DAKT. Firstly, a weight-based initial population selection strategy is proposed to constrain the generation of the initial population, which can enable the KT model to learn the potential distribution of the population. Secondly, a deep model is used to train the KT model to improve the quality of transferred knowledge. The negative KT phenomenon between tasks is reduced through adaptive adjustment of both knowledge transfer intensity and knowledge generation strategies. Finally, an offspring generation strategy based on inter-task and intra-task knowledge is designed, which can generate high-quality offspring individuals and accelerate the convergence of the algorithm. Intensive experiments on various benchmark problems demonstrate that the proposed algorithm outperforms state-of-the-art approaches and successfully obtain better convergence performance.
Keywords:
Evolutionary computation
evolutionary multitasking
multi-objective multi-task optimization
generative adversarial networks (GANs)
transfer adaptation
Journal
I
IF:
6.5
Papers:
1.4K
Citations:
4.5K
View more organizations