Scholarlink

This paper addresses a multiprocessor scheduling problem with real-world-inspired constraints. A learning-based prediction approach is proposed to support an effective algorithm selection during problem-solving. Classifiers are trained to predict whether a particular solver should be employed. The core of the contribution is the selection and engineering of features based on the structural properties of problem instances. These features are used as predictors in several well-known binary classification approaches. Computational experiments based on randomly generated problem instances reveal accuracy scores of up to 90% and indicate the potential time savings when embedding the proposed classifiers into the overall solution procedure.

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

Learning-Based Algorithm Selection for a Multiprocessor Scheduling Problem

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Roland Braune

摘要

This paper addresses a multiprocessor scheduling problem with real-world-inspired constraints. A learning-based prediction approach is proposed to support an effective algorithm selection during problem-solving. Classifiers are trained to predict whether a particular solver should be employed. The core of the contribution is the selection and engineering of features based on the structural properties of problem instances. These features are used as predictors in several well-known binary classification approaches. Computational experiments based on randomly generated problem instances reveal accuracy scores of up to 90% and indicate the potential time savings when embedding the proposed classifiers into the overall solution procedure.