基于进化优化的神经架构搜索方法综述  

A Survey of Evolutionary Optimization-based Neural Ar-chitecture Search

作　　者：赵海童 杨宇飞 ZHAO Haitong;YANG Yufei(School of Computer Science and Engineering,Changshu Institute of Technology,Changshu 215500;Software College of Northeastern University,Liaoning 110819,China)

机构地区：[1]常熟理工学院计算机科学与工程学院,江苏常熟215500 [2]东北大学软件学院,沈阳110819

出　　处：《常熟理工学院学报》2024年第5期1-14,共14页Journal of Changshu Institute of Technology

基　　金：国家自然科学基金项目“基于异构数据表征学习的超多任务多目标进化优化算法研究”(62302064)。

摘　　要：探讨了基于进化优化的神经架构搜索(Neural Architecture Search,NAS)算法,该类算法通过模拟生物进化的过程,能自动发现适应于特定任务的神经网络结构.本文回顾了进化优化算法在全局搜索、适应性和对离散连续问题的灵活性等方面的优势,详细介绍了基于进化计算的NAS方法在神经网络设计中的应用,强调了其在非连续、高维搜索空间中的潜力,总结了该领域的主要进展和挑战.最后,基于进化优化的NAS算法在推动神经网络架构搜索领域取得更多突破进行展望.未来的研究方向包括提高算法的计算效率、引入智能的搜索策略、克服高维搜索空间的问题以及该算法与深度学习的结合.This paper explores the evolutionary computation(EC)-based neural architecture search(NAS)algorithms,which can automatically discover neural network structures suitable for specific tasks by simulating the process of biological evolution.This paper reviews the advantages of EC-based NAS algorithms in global search,adaptability and flexibility for both continuous and discrete optimization problems.The application of EC in NAS,especially in consideration of its potential in searching in discrete and high-dimensional space,is introduced in detail.Furthermore,this paper summarizes the main advances and challenges in EC-based NAS algorithms.Finally,this paper looks forward to the prospect of EC-based NAS algorithms in driving more breakthroughs in the field of neural network architecture search.Future research directions include improvement of the computa-tional efficiency of the algorithms,introduction of intelligent search strategies,solution of the problem of high-dimensional search space,and the integration with deep learning.

关 键 词：人工智能 自动机器学习 进化计算 神经架构搜索 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]