Scholarlink

<p>The interpretation of neural networks is a hot research topic. An interpretable model for RBF neural networks is proposed from the perspective of rough sets, and it has better interpretation after combining the rules of the neural network with the attribute significance of rough sets. Firstly, the rule integration and extraction algorithm of neurons is proposed on the basis of the RBF hidden layer neuron influence ability. Secondly, the Pearson correlation coefficient is improved and the interpretation factor <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13042_2024_2522_Article_IEq1.gif" Format="GIF" Height="12" Rendition="HTML" Resolution="72" Type="Linedraw" Width="13" /> </InlineMediaObject> <EquationSource Format="TEX">\(\gamma\)</EquationSource> <EquationSource Format="MATHML"><math> <mi>γ</mi> </math></EquationSource> </InlineEquation> is defined, so the numerical evaluation of the interpretation of a neuron is carried out. An optimization algorithm for RBF neural networks is proposed. Finally, the feasibility of the interpretation model is analyzed through two sets of experiments, including interpretability experiments and optimization experiments. Experimental results in nine UCI datasets show that the proposed model has effective interpretation and can optimize the performance of RBF neural networks. The results provide heuristic information for research on the combination of rough sets with machine learning and the interpretation of neural networks.</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

Rough set interpretation to RBF neural network

  • 期刊论文

  • 发表日期：

    2025年1月9日

查看全文
  • Dayong Deng,
  • Jie Wang,
  • Zhixuan Deng,
  • Keyu Liu,
  • Pengfei Zhang

摘要

The interpretation of neural networks is a hot research topic. An interpretable model for RBF neural networks is proposed from the perspective of rough sets, and it has better interpretation after combining the rules of the neural network with the attribute significance of rough sets. Firstly, the rule integration and extraction algorithm of neurons is proposed on the basis of the RBF hidden layer neuron influence ability. Secondly, the Pearson correlation coefficient is improved and the interpretation factor \(\gamma\) γ is defined, so the numerical evaluation of the interpretation of a neuron is carried out. An optimization algorithm for RBF neural networks is proposed. Finally, the feasibility of the interpretation model is analyzed through two sets of experiments, including interpretability experiments and optimization experiments. Experimental results in nine UCI datasets show that the proposed model has effective interpretation and can optimize the performance of RBF neural networks. The results provide heuristic information for research on the combination of rough sets with machine learning and the interpretation of neural networks.