obilogy / G3NS

Optimization of an Up-scalable Graph Neural Network based on the C. Elegans Connectome Nervous System

Optimization of an Up-scalable Graph Neural Network based on the C. Elegans Connectome Nervous System (G3NS)

Introduction

G3NS's ideas lie at the intersection of biology and bio-inspired machine learning, with a focus on C. Elegans.

Latest Update: 4/9/2024 (see below)

Background

• C. Elegans Connectome: This organism's neural network, though intricate, is the most detailed connectome available to researchers. C. Elegans Connectome edge list modified from the work of Ripoll-Sanchez et al..
• Understanding Neural Dynamics: The capability of biological neural connections to not only process and respond to stimuli, but also maintain a power efficient, plastic brain state remains ellusive.
• Beyond Traditional ANNs: Conventional Artificial Neural Networks may fall short in mimicking the intricate web of biological neural networks, driving the need for bio-inspired approaches.
• Broader Implications: Particularly exciting, is the idead that all mammal connectomes follow similar operational principles, thus the realization of this project could offer clues applicable to more complex neural systems.

Potential Project Stages

1. Modeling G3NS: Crafting a neural network model inspired by the C. Elegans connectome.
2. Environmental Simulation: Testing the model's response to sensory inputs within a simulated environment.
3. Reinforcement Learning Integration: Employing techniques like Proximal Policy Optimization (PPO) or Heterogeneous Graph Neural Architecture Search (GHGNAS) to optimize the functions driving the network's learning.
4. Neural Pattern Analysis: In-depth analysis of simulated neural activities. These are tensorboard images from the first time I ran the code.
   • 
   • 
   • 
   • 

UPDATE:

This has become a dormant project for now. However, for those of you who might be interested in related research please see:

MetaWorm: A Complete Model Bridging Brain, Body and Environment of C. elegans

Mengdi Zhao, Ning Wang, Xinrui Jiang, Xiaoyang Ma, Haixin Ma, Gan He, Kai Du, Lei Ma and Tiejun Huang

bioRxiv. posted 26 February 2024, 10.1101/2024.02.22.581686

http://biorxiv.org/content/early/2024/02/26/2024.02.22.581686

---