Question about the internal neuron from the famous youtube video

[piotrkolodziej]

I have a question about the internal N0 neuron from the following wiring. Can it even be computed?

[davidrmiller]

Hi @piotrkolodziej, yes it can. In each simulator step, every neuron is evaluated and each one latches it output until the next sim step. In each sim step, we start by evaluting each input neuron, then each inner neuron, then each output neuron. When a neuron is evaluated, each of its inputs comes from whatever was most recently latched in the output of the other neuron or its own previously latched output if it feeds itself. When an inner neuron has only one input from its own output, like N0, then its behavior over the course of its lifetime depends on the connection weight, which comes from its genome, and its initial output value set at birth, which is set by the function initialNeuronOutput(). I believe that, depending on those values, it could eventually converge to a constant output value, or could oscillate between positive and negative values.

[piotrkolodziej]

@davidrmiller than you for the response. I hope you don't mind picking your brain a little further as I have a follow up question.

I like the simplicity of your approach (processing fist input, then inner, then output). I am wondering though why not just traverse the tree from the 'bottom'? It seems that latching values wouldn't be needed and there would be no issues with the order of the inner neurons (I suppose with your approach, using the above picture as the example, processing N2 before N0 is not possible unless you assume some value between N0-N2).

[davidrmiller]

That's an interesting question. It would be interesting to explore other ways to update the network. I confess I just went with what seemed intuitive at the time without considering the pros and cons of other methods.