新颖性
计算机科学
元启发式
方向(向量空间)
线性判别分析
判别式
人工智能
算法
模式识别(心理学)
机器学习
数学
心理学
社会心理学
几何学
作者
Zhongbo Hu,Qian Zhang,Yujie Wang,Qinghua Su,Zenggang Xiong
标识
DOI:10.1016/j.asoc.2024.111521
摘要
The rapid rate of generating a new metaheuristic algorithm almost every month is causing increasing concerns and disputes about their novelty. To stop the disputes and steer algorithm design in a healthy direction, this article presents a discriminant method of novelty and a research orientation for metaheuristic algorithms. The novelty discriminant is implemented by two novel mathematical definitions of homologous algorithms and root algorithms. The two definitions are developed to divide algorithms into two classes according to a discrepancy in whether the reproduction operator of an algorithm is a linear combination of existing operators. Root algorithms are strongly innovative because of the novelty of their reproduction operators. A homologous algorithm is recognized as a novel algorithm only when the practical value and academic significance of the new combinatorial structure of the algorithm's reproduction operator is clearly highlighted. So a research orientation that the study of a homologous algorithm should focus on how a certain metaphor evokes a new combinatorial structure can be developed. Moreover, numerical experiments should be conducted to analyze the relationship between its search behavior and its new combinatorial structure. Further work can be directed towards studying the systematization of existing knowledge about search behaviors of metaheuristic algorithms.
科研通智能强力驱动
Strongly Powered by AbleSci AI