Scholarlink

Building upon our previous work on integrating Cultural Algorithms (CA) with the Policy Gradients with Parameter-Based Exploration (PGPE) algorithm, this study presents an enhanced version of the CA-PGPE algorithm for the MNIST hand-written digit classification task. The improved CA-PGPE incorporates Topographic knowledge source (KS) and refined knowledge source weighting to efficiently navigate the search space and improve convergence speed. By leveraging the belief space consisting of Domain, Situational, History, Topographic, and Normative knowledge sources, the enhanced CA-PGPE algorithm demonstrates improved performance compared to the original PGPE algorithm. The results showcase the potential of integrating cultural knowledge into evolutionary optimization algorithms for tackling complex machine learning tasks. Future research directions include exploring the application of CA-PGPE to evolve InfoGAN network parameters and investigating the implementation of social learning mechanism within the population space.

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

Enhancing Cultural Algorithm Guided Policy Gradients with Parameter-Based Exploration Through Topographic Knowledge and Adaptive Weighting

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mark Nuppnau,
  • Khalid Kattan,
  • R. G. Reynolds

摘要

Building upon our previous work on integrating Cultural Algorithms (CA) with the Policy Gradients with Parameter-Based Exploration (PGPE) algorithm, this study presents an enhanced version of the CA-PGPE algorithm for the MNIST hand-written digit classification task. The improved CA-PGPE incorporates Topographic knowledge source (KS) and refined knowledge source weighting to efficiently navigate the search space and improve convergence speed. By leveraging the belief space consisting of Domain, Situational, History, Topographic, and Normative knowledge sources, the enhanced CA-PGPE algorithm demonstrates improved performance compared to the original PGPE algorithm. The results showcase the potential of integrating cultural knowledge into evolutionary optimization algorithms for tackling complex machine learning tasks. Future research directions include exploring the application of CA-PGPE to evolve InfoGAN network parameters and investigating the implementation of social learning mechanism within the population space.