Scholarlink

<p>Fuzzy multi-view modeling inevitably involves the following challenges: 1) effectively utilizing the shared decision information among views to facilitate the training of fuzzy systems and 2) optimizing the fuzzy training structure through mutual approximation among views. To this end, a two-view mutual approximation fuzzy classifier (Sdivf-FC) is proposed herein. First, by integrating the knowledge decision matrices of each view, a joint-view shared decision subspace is constructed, which guides the decision-making process of each single view in every layer. This design enhances the learning efficiency of single views and maximizes the utilization of shared information among views. Second, a two-view mutual approximation–based optimization strategy is proposed to optimize the consequent parameters of fuzzy rules through intra-view and inter-view approximations. Additionally, a novel stacked-like structure is designed, which combines the projection of the previous layer’s input space and the prediction errors of the two views to generate a new input space, thereby accelerating the training process and increasing the diversity of training samples. To maintain the interpretability of Sdivf-FC, an antecedent parameter inheritance mechanism is also proposed. Experimental results demonstrate that Sdivf-FC outperforms similar classifiers on both benchmark UCI datasets and epilepsy electroencephalogram datasets.</p>

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 Hierarchical-approximation-based Interpretable Multi-view TSK Fuzzy Classification for Diverse Aggregates of Fuzzy Rules

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Jia Zhai,
  • Jinghao Chen,
  • Guanghao Song,
  • Yuanqing Yang,
  • Wei Yan,
  • Ta Zhou

摘要

Fuzzy multi-view modeling inevitably involves the following challenges: 1) effectively utilizing the shared decision information among views to facilitate the training of fuzzy systems and 2) optimizing the fuzzy training structure through mutual approximation among views. To this end, a two-view mutual approximation fuzzy classifier (Sdivf-FC) is proposed herein. First, by integrating the knowledge decision matrices of each view, a joint-view shared decision subspace is constructed, which guides the decision-making process of each single view in every layer. This design enhances the learning efficiency of single views and maximizes the utilization of shared information among views. Second, a two-view mutual approximation–based optimization strategy is proposed to optimize the consequent parameters of fuzzy rules through intra-view and inter-view approximations. Additionally, a novel stacked-like structure is designed, which combines the projection of the previous layer’s input space and the prediction errors of the two views to generate a new input space, thereby accelerating the training process and increasing the diversity of training samples. To maintain the interpretability of Sdivf-FC, an antecedent parameter inheritance mechanism is also proposed. Experimental results demonstrate that Sdivf-FC outperforms similar classifiers on both benchmark UCI datasets and epilepsy electroencephalogram datasets.