《计算机应用》唯一官方网站

所属专题: 进化计算专题(2024年第5期“进化计算专题”导读,全文即将上线)

• •    下一篇

多任务优化算法及应用研究综述

武越1,丁航奇1,何昊1,毕顺杰1,江君1,公茂果2,苗启广3,马文萍1   

  1. 1. 西安电子科技大学 计算机科学与技术学院,西安 7100712. 西安电子科技大学 电子工程学院,西安 7100713. 西安电子科技大学 人工智能学院,西安 710071

  • 收稿日期:2024-03-04 发布日期:2024-04-26 出版日期:2024-04-26
  • 通讯作者: 公茂果
  • 作者简介::武越(1988—),男,陕西西安人,副教授,博士,CCF高级会员,主要研究方向:人工智能、三维视觉; 丁航奇(1998—),男,山东烟 台人,博士研究生,CCF会员,主要研究方向:进化计算、点云配准; 何昊(2002—),男,江西南昌人,硕士研究生,CCF会员,主要研究方向:进化 计算、点云配准; 毕顺杰(2000—),男,山东菏泽人,硕士研究生,CCF会员,主要研究方向:进化计算、点云配准; 江君(2002—),男,江西上饶 人,硕士研究生,CCF会员,主要研究方向:进化计算、点云配准; 公茂果(1979—),男,山东临沂人,教授,博士,CCF高级会员,主要研究方向: 人工智能; 苗启广(1972—),男,山东青岛人,教授,博士,CCF杰出会员,主要研究方向:计算机视觉、大数据分析; 马文萍(1981—),女,陕西 西安人,教授,博士,CCF专业会员,主要研究方向:自然计算、智能图像处理。
  • 基金资助:
    国家自然科学基金资助项目(6203600662276200);中国人工智能学会-华为MINDSPORE学术奖励基金资助项目。

Research review of multitask optimization algorithms and applications

WU Yue1DING Hangqi1HE Hao1BI Shunjie1JIANG Jun1GONG Maoguo2*MIAO Qiguang1MA Wenping3   

  1. 1. School of Computer Science and TechnologyXidian UniversityXian Shaanxi 710071China2. School of Electronic EngineeringXidian UniversityXian Shaanxi 710071China3. School of Artificial IntelligenceXidian UniversityXian Shaanxi 710071China
  • Received:2024-03-04 Online:2024-04-26 Published:2024-04-26
  • Contact: Maoguo Gong
  • About author:WU Yue,born in 1988,Ph. D.,associate professor. His research interests include artificial intelligence,3D vision. DING Hangqi,born in 1998,Ph. D. candidate. His research interests include evolutionary computation,point cloud registration. HE Hao,born in 2002,M. S. candidate. His research interests include evolutionary computation,point cloud registration. BI Shunjie,born in 2000,M. S. candidate. His research interests include evolutionary computation,point cloud registration. JIANG Jun,born in 2002,M. S. candidate. His research interests include evolutionary computation,point cloud registration. GONG Maoguo,born in 1979,Ph. D.,professor. His research interests include artificial intelligence. MIAO Qiguang,born in 1972,Ph. D.,professor. His research interests include computer vision,big data analysis. MA Wenping,born in 1981,Ph. D.,professor. Her research interests include natural computing,intelligent image processing.
  • Supported by:
    This work is partially supported by National Natural Science Foundation of China 6227620062036006) , CAAI-Huawei MINDSPORE Academic Open Fund.

摘要: 进化多任务优化(EMTO)是进化计算中一种新型的方法之一,它可以同时解决多个相关的优化任务,并通过任务之间的知识转移来增强每个任务的优化。近年来,越来越多的进化多任务相关研究致力于利用它强大的并行搜索能力和降低计算成本的潜力来优化各种问题,并且将EMTO应用于各种各样的实际场景。从EMTO的原理、核心设计、应用以及挑战四个方面对EMTO的算法及应用进行了讨论。首先介绍了EMTO的大致分类,分别从两个层次、四个方面介绍,包括单种群多任务、多种群多任务、辅助任务形式以及多形式任务形式;其次介绍EMTO的核心组件设计,包括任务构建以及知识转移,最后对它的各种应用场景进行介绍,并对今后研究做了总结与展望。

关键词: 进化多任务优化(EMTO), 单种群多任务, 多种群多任务, 多形式任务, 知识转移

Abstract: Evolutionary Multitask Optimization (EMTO) is one of the new methods in evolutionary computing, which can simultaneously solve multiple related optimization tasks and enhance the optimization of each task through knowledge transfer between tasks. In recent years, more and more research on evolutionary multitasking has been devoted to utilizing its powerful parallel search capabilities and potential to reduce computational costs to optimize various problems, and applying EMTO to various practical scenarios. The research and application of EMTO were discussed from four aspects: its principle, core design, application, and challenges. Firstly, the general classification of EMTO was introduced from two levels and four aspects, including single-population multitask, multipopulation multitask, auxiliary task forms, and multiform task; Next, we will introduce the core component design of EMTO, including task construction and knowledge transfer. Finally, we will introduce its various application scenarios and provide a summary and outlook for future research.

Key words: Evolutionary multitask optimization (EMTO), Single-population multitasking, Multipopulation multitasking, Multiform task, Knowledge transfer