Adaptive Estimation Distribution Multitask Optimization

  • Zhou, Shi-Yuan; 
  • Wang, Zi-Jia; 
  • Yu, Wei-Jie; 
  • Zhao, Tian-Fang; 
  • Zhan, Zhi-Hui; 
  • ... Zhang, Jun; 
  • 외 1명
Citations

WEB OF SCIENCE

2
Citations

SCOPUS

0

초록

Evolutionary multitask optimization (EMTO) requires optimizing multiple tasks simultaneously by knowledge transfer (KT). However, the current KT techniques often have two limitations. First, most KT techniques do not fully exploit the distribution or similarity information of tasks from multiple aspects. Since the KT efficiency is strongly related to multiple aspects of the similarity and distribution information in tasks, if the distribution information can be fully mined and utilized, the KT process will be effectively guided. Second, current KT methods usually only consider the information exchange on the index-aligned dimensions and ignore the information exchange on other index-unaligned dimensions. However, the index-aligned dimensions are not always suitable for information communication. Therefore, in this article, an adaptive estimation distribution multitask optimization (AEDMTO) algorithm is proposed to overcome these two limitations. For the first limitation, we propose an adaptive estimation distribution (AED) strategy to fully estimate the distribution and evaluate the similarity of tasks from both individual and population aspects, so as to obtain the useful transfer information and adaptively adjust the KT frequency. For the second limitation, a stochastic dimension-matching (SDM) strategy is introduced to realize KT on the index-unaligned dimensions, which helps to match more suitable dimensions for KT in different tasks. AEDMTO shows its significant superiority when compared with other state-of-the-art multitask algorithms on two multitask test suites, CEC2017 and CEC2022. Moreover, AEDMTO is applied to a real-world multitask sensor coverage problem (SCP) to reveal its applicability. Finally, the extended experiments on many-task optimization problems illustrate the scalability of AEDMTO.

키워드

Optimization; Optimization; Multitasking; Multitasking; Algorithms; Algorithms; Frequency; Frequency; Indexes; Indexing; Indexing; Image sensors; Image sensors; Materials; Materials; Testing; Testing; Modeling; Modeling; Adaptive estimation distribution (AED); Adaptive estimation distribution (AED); evolutionary multitask optimization (EMTO); evolutionary multitask optimization (EMTO); knowledge transfer (KT); knowledge transfer (KT); MULTIOBJECTIVE EVOLUTIONARY ALGORITHM; VEHICLE-ROUTING PROBLEM
제목
Adaptive Estimation Distribution Multitask Optimization
저자
Zhou, Shi-Yuan; Wang, Zi-Jia; Yu, Wei-Jie; Zhao, Tian-Fang; Zhan, Zhi-Hui; Kwong, Sam; Zhang, Jun
DOI
10.1109/TCYB.2026.3741167
발행일
2026-10
유형
Article; Early Access
저널명
IEEE Transactions on Cybernetics