Neural Model
- For this experiment, we (MDX) used our old simulator. We should
be able to put it in Nest and SpiNNaker.
- We're the only group I'm aware of exploring this.
- I've spoken about our FLIF model
before. So look here for more information.
- The simulation uses a fatiguing integrate and fire (FLIF) neuron model.
- The model is a specialisation of integrate and fire. So they're spiking.
- It works in discrete cycles.
- Integration is done by combining inputs and if a neuron passes a
threshold it fires.
- If the neuron does not fire, it loses some of its activation;
it leaks away. The retained activity makes it easier to fire
in the next cycle.
- Each neuron also has a fatigue level. If it fires fatigue increase,
and if it doesn't fatigue decreases (to at most 0).
- Fatigue increases the threshold.
- Earlier these parameters were set for some biological data with
theta = 2.2, d = 1.12, Fatigue = 0.045 and Fatigue Recovery at
0.01.
- On a widely varying input regime (via electrodes), this gets
about 90% of spikes right (see
the IMCIC talk for more).
