Scholarlink

The paper considers a parallel algorithm for solving global optimization problems with a partially defined objective function of the “black box” type. Such problems arise in the process of selecting optimal values of hyperparameters used in machine learning and artificial intelligence methods. The algorithm employs a dimensionality reduction scheme based on Peano curves in combination with an asynchronous scheme of parallel computing. To demonstrate the effectiveness of parallelization, computational experiments were conducted for the task of tuning the hyperparameters of artificial intelligence methods used to predict the values of a time series.

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

A Parallel Algorithm for Solving Global Optimization Problems and Its Application for Tuning Hyperparameters of AI Methods

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Marina Usova,
  • Ilya Lebedev,
  • Anton Shtanyuk,
  • Evgeny Kozinov,
  • Konstantin Barkalov

摘要

The paper considers a parallel algorithm for solving global optimization problems with a partially defined objective function of the “black box” type. Such problems arise in the process of selecting optimal values of hyperparameters used in machine learning and artificial intelligence methods. The algorithm employs a dimensionality reduction scheme based on Peano curves in combination with an asynchronous scheme of parallel computing. To demonstrate the effectiveness of parallelization, computational experiments were conducted for the task of tuning the hyperparameters of artificial intelligence methods used to predict the values of a time series.