Neuron Models

• There are a lot of different models.
• There are several simple integrate and fire models (like the perceptron). Integrate and fire neurons integrate activity from other neurons and if it goes over the threshold, it fires.
• It's like a binary perceptron, but it functions in time, so can fire multiple times. This is called the McCullouch Pitts neuron.
• Integration is done by combining inputs and if a neuron passes a threshold it fires.
• You can extend this so some of the activity leaks away.
• 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.
  
• It can work in discrete cycles (which is the way they usually go), or continuously.
• I extend this to fatiguing leaky integrate and fire models.
• Other examples are Boltzmann machines and Izhikevich neurons.
• These are all point models, but you can use more sophisticated models of neurons using their shape and conductance.
• These are called compartmental models because they break the neuron into compartments and model how the current moves about them. They're more accurate than point models.
• A Nobel prize winning model is from Hodgkin and Huxley, who use ion channels.