Neurons, as in the brain, are the base “processing” unit of DN. Unlike most neural networks in existence today, DN’s neurons look like this:

Rather than this:

DN takes the approach of modeling real biological processes. This means the neurons in DN have bodies, axons and dendrites. They are somewhat crude approximations in that they are simplified but they are far more flexible than the current artificial neural nets being used.

Neurons in DN have three dimensional positions. When the network is initially grown, neurons are placed randomly. The number of neurons and physical size of the network are specified in a configuration file (which can be modified as the evolution process continues). Each neuron is given a base set of parameters. These parameters define things such as dendrite growth direction and distance, maximum potential before firing, time required to fire and recover, etc. Initially all neurons are given the same parameters as defined as the configuration file.

Each neuron, however, stores these values independantly and can modify them. The modification process itself is evolved. Honestly, I don’t (nor do any neuroscientists) have a good idea how the self-potentiation process occurs. So I leave it up to evolution. Receptors on a neuron’s dendrite can polarize/depolarize the neuron and force it to fire. Alternatively, they can be randomly assigned a “pathway”. These pathways have various traits that modify the parameters of the neuron. These modifications can be temporary or permanent but are stored for that individual neuron. One neuron may respond to neurotransmitter A by depolarizing while another neuron responds to the same neurotransmitter by triggering a pathway to shorten the recovery time before firing.

The evolution simulation will take several neural genotypes, grow the networks and then run them through the simulation many times. After x many iterations, a particular genotype will be evaluated to see how fit it is (how well it survives and how plastic it is). Genotypes are then discarded or kept depending on their fitness score. Highly fit networks will have their parameters adjusted by a small random amount and then regrown. These regrown networks will have similar parameters to their parents but different physical layouts, as well as different individual neuronal configurations as the network modifies itself through the simulation.