Scholarlink

<p>A novel MorphoNAS approach is proposed for deterministic neural network growth through morphogen-guided self-organization based on the free energy principle, reaction-diffusion systems, and gene regulatory networks. The developmental model is considered, where compact genomes encode only morphogen dynamics and threshold-based cellular rules enabling single progenitor cell transformation into complex neural architectures. Full success (100%) is achieved in evolutionary search for genomes generating predefined graph configurations with 8–31 nodes. Minimal functional controllers (6–7 neurons) for the CartPole problem are obtained under network size minimization pressure with 94% population success rate. The results demonstrate that biologically plausible developmental rules can serve as an effective mechanism for automated neural architecture search.</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

MorphoNAS: Embryogenic Neural Architecture Search Through Morphogen-guided Development

  • 期刊论文

  • 发表日期：

    2026年3月1日

查看全文
  • M. M. Glybovets,
  • S. O. Medvid

摘要

A novel MorphoNAS approach is proposed for deterministic neural network growth through morphogen-guided self-organization based on the free energy principle, reaction-diffusion systems, and gene regulatory networks. The developmental model is considered, where compact genomes encode only morphogen dynamics and threshold-based cellular rules enabling single progenitor cell transformation into complex neural architectures. Full success (100%) is achieved in evolutionary search for genomes generating predefined graph configurations with 8–31 nodes. Minimal functional controllers (6–7 neurons) for the CartPole problem are obtained under network size minimization pressure with 94% population success rate. The results demonstrate that biologically plausible developmental rules can serve as an effective mechanism for automated neural architecture search.