Scholarlink

<p>This paper proposes Mycelium-NAS, a novel, biologically inspired neural architecture search (NAS) approach based on mycelium network growth and adaptation. This novel approach formulates NAS as an emergent growth process, where architectures dynamically evolve through adaptive expansion, fusion, and reorientation of network components, mirroring the organic exploration of mycelium. In particular, it introduces hyphal growth strategies–branching, fusion, and orientation–as new operations for dynamically constructing neural architectures. A fitness-driven selection process optimizes performance while penalizing excessive parameters. Evaluated on image classification tasks, Mycelium-NAS demonstrates competitive performance on CIFAR-10, CIFAR-100, and Fashion-MNIST datasets, achieving a compelling balance between accuracy and parameter efficiency in comparison to established architectures. Our results highlight the potential of biologically-inspired search strategies to address the critical challenges in NAS, offering a resource-conscious and adaptable paradigm for automated neural network design.</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

Mycelium neural architecture search

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Tanvir Islam

摘要

This paper proposes Mycelium-NAS, a novel, biologically inspired neural architecture search (NAS) approach based on mycelium network growth and adaptation. This novel approach formulates NAS as an emergent growth process, where architectures dynamically evolve through adaptive expansion, fusion, and reorientation of network components, mirroring the organic exploration of mycelium. In particular, it introduces hyphal growth strategies–branching, fusion, and orientation–as new operations for dynamically constructing neural architectures. A fitness-driven selection process optimizes performance while penalizing excessive parameters. Evaluated on image classification tasks, Mycelium-NAS demonstrates competitive performance on CIFAR-10, CIFAR-100, and Fashion-MNIST datasets, achieving a compelling balance between accuracy and parameter efficiency in comparison to established architectures. Our results highlight the potential of biologically-inspired search strategies to address the critical challenges in NAS, offering a resource-conscious and adaptable paradigm for automated neural network design.