@El'endiaStarman so basically what I've been reading up on are the different models for neurons.
Models range from full-blown neuron simulations (dozens of parameters, physically realistic, etc.) to really simple (1 = firing, 0 = not firing).
The larger models are too computation-heavy to use for big simulations (one paper says 1200 floating-point operations per neuron per millisecond).
And so basically the question is, how simple can we make the model but still have the neuron act realistic. (Turns out, neurons can fire in all sorts of patterns, the paper gives some illustrations on page 2).
And so this author presents one model that can do pretty much all of those behaviors, but in which the neuron's state has 2 variables, there are 4 parameters, and which works well for time-stepping.
For a golf challenge I would have like a single neuron with a fixed input current.
This model takes like 20 floating point ops per neuron per millisecond.
Also, I just had a weird moment of self-awareness - I'm reading a paper in a field that's not my own for fun over breakfast on a Sunday. That's something very very few people would do.
@PhiNotPi: So in the expression $v' = 0.04v^2 + 5v + 140 - u + I$, that $I$ can be the output $v$ of another neuron that's coupled to it? If so, how do you handle multiple neurons coupled to one neuron?
@El'endiaStarman I haven't really looked at those details yet. Yes, you would sum the inputs, but there is probably variation in how the input of one neuron is related to the output of another.
Like, synapses can have different strengths of connection between two neurons.
When you place one spinner inside of the other, you tend to get non-chaotic patterns, and I'm thinking there could be wallpaper potential in some of them.
