Integrate-and-fire

The integrate-and-fire model of neural networks followings on nicely from my last post and also ties in with the subject of the next chapter. It's a phenomenological model of neurons which "integrates" the depolarizations incident on a neuron's cell body and "fires" an action potential down the axon when the depolarization reaches the threshold value. Often additional time dependent exponential decay terms are added to better replicate observed behaviour; when calibrated carefully this model is a cheap and robust method of investigating neural networks.

While integrate-and-fire doesn't allow for the investigation of temporal encoding, it can be used to model large neural networks and captures some of the effects of combining analogue and digital signals along with the topological structure in the network. It is a model liked by many computational neuroscientists and disliked by biologists a physicists in equal measure (just for different reasons).

For more details see: A review of the integrate-and-fire neuron model: II. Inhomogeneous synaptic input and network properties.