Scholarlink

Parallel computing has become popular since multi-core advancements followed by multi-core architectures such as GPUs. Thus, it is essential to leverage these technologies when developing new applications to achieve better performance. In this context, this work investigates the parallel computing capabilities of TensorFlow to solve eight well-known optimization problems by applying metaheuristics such as Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, and Cuckoo Search. Results show that TensorFlow can achieve a speedup of up to eight times in high-dimension benchmark functions with low programming effort using only one GPU. This work contributes to the body of knowledge by demonstrating and measuring how to get benefits from TensorFlow in metaheuristic implementations.

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

Performance Evaluation of Metaheuristics Using TensorFlow Parallelism on GPU

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Gustavo Gusmão Rocha,
  • Fábio M. F. Lobato,
  • Omar Andres Carmona Cortes

摘要

Parallel computing has become popular since multi-core advancements followed by multi-core architectures such as GPUs. Thus, it is essential to leverage these technologies when developing new applications to achieve better performance. In this context, this work investigates the parallel computing capabilities of TensorFlow to solve eight well-known optimization problems by applying metaheuristics such as Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, and Cuckoo Search. Results show that TensorFlow can achieve a speedup of up to eight times in high-dimension benchmark functions with low programming effort using only one GPU. This work contributes to the body of knowledge by demonstrating and measuring how to get benefits from TensorFlow in metaheuristic implementations.