Scholarlink

This paper presents a novel approach to enhance the performance of Radial Basis Function Neural Networks (RBFNNs) for regression tasks through an innovative integration of Markov Chain Monte Carlo (MCMC) sampling with Fuzzy C-Means clustering. Traditional RBFNNs, while effective for various pattern recognition tasks, often face challenges in optimal placement of their radial basis functions. We address this limitation by introducing a hybrid methodology that combines the robust exploration capabilities of MCMC with the flexibility of fuzzy clustering. Our approach dynamically adjusts RBF centers based on both data distribution and regression performance metrics, utilizing cross-entropy loss as a key indicator for optimization. The proposed framework incorporates a comprehensive learning strategy that includes strategic initialization of RBF centers, adaptive parameter adjustment, and weighted least square error estimation for coefficient optimization. Experimental results across multiple benchmark datasets demonstrate significant improvements in regression tasks compared to conventional RBFNNs implementations, while maintaining computational efficiency.

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 Novel Design of Radial Basis Function Neural Network Integrating Markov Chain Monte Carlo Clustering Algorithm

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yunlong Zhu,
  • Zunwei Fu,
  • Zheng Wang

摘要

This paper presents a novel approach to enhance the performance of Radial Basis Function Neural Networks (RBFNNs) for regression tasks through an innovative integration of Markov Chain Monte Carlo (MCMC) sampling with Fuzzy C-Means clustering. Traditional RBFNNs, while effective for various pattern recognition tasks, often face challenges in optimal placement of their radial basis functions. We address this limitation by introducing a hybrid methodology that combines the robust exploration capabilities of MCMC with the flexibility of fuzzy clustering. Our approach dynamically adjusts RBF centers based on both data distribution and regression performance metrics, utilizing cross-entropy loss as a key indicator for optimization. The proposed framework incorporates a comprehensive learning strategy that includes strategic initialization of RBF centers, adaptive parameter adjustment, and weighted least square error estimation for coefficient optimization. Experimental results across multiple benchmark datasets demonstrate significant improvements in regression tasks compared to conventional RBFNNs implementations, while maintaining computational efficiency.