FLIF Neurons
- I use a fatiguing leaky integrate and fire model.
- It uses discrete cycles.
- It integrates activity from connected firing neurons, and fires
if it passes a threshold.
- If it doesn't fire, some but not all of the activity leaks away.
- If it does fire, it accumulates fatigue (losing if it doesn't fire).
- This raises the threshold making it harder to fire.
- I aligned it to some neural data and it predicts spikes
within 2 cycles around 90% of the time.
- My cycles are 10 ms. long, so I can ignore synaptic
delay and refractory periods.
- It's also pretty efficient so I can manage 100,000 neurons on a PC
in real time.
- Modellers and tooth brushes.
