Scholarlink

In a typical neural network, the neurons are real-valued or complex valued. In some recent work, the notion of neurons being (closed) interval-valued has been developed. Such a formulation can be useful in applications where data are interval-valued. In this paper, we develop the idea of convex-polygon-valued neurons. When exploring examples of these two types of networks, we also include sparsity and entropy criteria, which are conflicting, with the goal of demonstrating that including such criteria with appropriately small coefficients can improve network performance.

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

From Interval-Valued Neurons to Convex-Polygon-Valued Neurons, with Sparsity and Entropy Criteria

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • B. Boreland,
  • H. Kunze,
  • K. Levere

摘要

In a typical neural network, the neurons are real-valued or complex valued. In some recent work, the notion of neurons being (closed) interval-valued has been developed. Such a formulation can be useful in applications where data are interval-valued. In this paper, we develop the idea of convex-polygon-valued neurons. When exploring examples of these two types of networks, we also include sparsity and entropy criteria, which are conflicting, with the goal of demonstrating that including such criteria with appropriately small coefficients can improve network performance.