Fatiguing, Leaky, Integrate, and Fire Neurons
- For a long time people have been simulating integrate and
fire neurons (Lapicque 1907; McCulloch and Pitts 1943).
- These neurons integrate activity from other neurons.
- If they pass a threshold, they fire and then pass activity
(weighted by the synaptic strength) to other neurons.
- We use a neural parameter for threshold.
- We, and a lot of others, use discrete cycles here, though
it can work in analog time either by time slicing aribitrarily
fine, or keeping event counters.
- What happens with activity that doesn't get used in a cycle?
- McCulloch and Pitts lose it; Lapicque keeps it all.
- In reality, some leaks away. This gives you the Leaky, Integrate
and Fire neuron. That's modeled with a decay factor that divides
the activity each cycle.
- Neurons also fatigue. If they fire a lot, they get harder to
fire. We model this with fatigue which is increased when
a neuron fires, and decreased when it doesn't. This uses
two constants.