Spontaneous Activation

It turns out that biological neurons fire without any input, typically about once a second.
We knew about this and did it a long time ago by just firing neurons randomly a percentage of the time.
This provides a mechanism to spread CAs beyond externally stimulated neurons. You can't increase a weight to a neuron unless it fires.
That was a bit unsatisfying, so we modified the fatigue model so fatigue could go below zero.
When it got to negative threshold the neuron would fire.
So, spontaneous firing now emerges naturally from the fatigue model.
In the biological data we were looking at there was no spontaneous firing, but the neuron fired more rapidly at first.
However, we do have to modify fatigue recovery when fatigue is negative, or we get too many spikes at first.
We use a fair few models including dividing the fatigue, but the best had an exponential recovery function.
I'm now looking into learning with this model.