Scholarlink

This paper presents an approach for the parallel launch of local optimization methods within a global search algorithm framework, aimed at solving global optimization problems. The objective function is treated as a black-box and is assumed that it satisfies the Lipschitz condition with an unknown constant. The proposed method focuses on selecting the region of attraction of local extrema of the objective function based on the analysis of accumulated search data. Employing machine learning techniques for this analysis enables informed decisions regarding the parallel execution of local methods, leading to improved algorithm convergence. The effectiveness of this approach is validated through numerical experiments, which demonstrate accelerated performance on a set of benchmark problems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Parallel Global Optimization Algorithm Employing Decision Trees for Launching Local Methods

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Konstantin Barkalov,
  • Ilya Lebedev,
  • Dmitry Silenko

摘要

This paper presents an approach for the parallel launch of local optimization methods within a global search algorithm framework, aimed at solving global optimization problems. The objective function is treated as a black-box and is assumed that it satisfies the Lipschitz condition with an unknown constant. The proposed method focuses on selecting the region of attraction of local extrema of the objective function based on the analysis of accumulated search data. Employing machine learning techniques for this analysis enables informed decisions regarding the parallel execution of local methods, leading to improved algorithm convergence. The effectiveness of this approach is validated through numerical experiments, which demonstrate accelerated performance on a set of benchmark problems.