Development and Evolution of Neural Networks in an Artificial Chemistry

TL;DR

This paper presents a decentralized, genome-driven framework for growing artificial neural networks via an artificial chemistry, integrating artificial physics, biochemistry, and autonomous gene expression. It demonstrates how diffusion-guided growth, cell-type signaling, and genome-defined behaviors can yield self-organizing, heterogeneous networks capable of simple associative learning, illustrated by a Pavlov-like conditioning example. To scale exploration, the authors implement a platform-independent, asynchronous distributed genetic algorithm that taps into Internet participation to evolve genomes. The approach offers a path toward more universal, developmentally grounded neurocomputational structures and scalable evolutionary search beyond hand-crafted designs.

Abstract

We present a model of decentralized growth for Artificial Neural Networks (ANNs) inspired by the development and the physiology of real nervous systems. In this model, each individual artificial neuron is an autonomous unit whose behavior is determined only by the genetic information it harbors and local concentrations of substrates modeled by a simple artificial chemistry. Gene expression is manifested as axon and dendrite growth, cell division and differentiation, substrate production and cell stimulation. We demonstrate the model's power with a hand-written genome that leads to the growth of a simple network which performs classical conditioning. To evolve more complex structures, we implemented a platform-independent, asynchronous, distributed Genetic Algorithm (GA) that allows users to participate in evolutionary experiments via the World Wide Web.

Figures (9)

• Figure 1: Hexagonal grid with boundary elements. Diffusion occurs from a local concentration peak at grid element $N$.
• Figure 2: Genetic structure of neural cells. Local concentrations of substrates (condition atoms) trigger gene expression.
• Figure 3: Development of the network for classical conditioning.
• Figure 4: Genome for development and behavior of network exhibiting classical conditioning
• Figure 5: A schematical representation of the network for classical conditioning. The types of neurotransmitter used are shown next to the axons. The cell-type protein used by each cell is indicated near the cell body.