Neural Models
- There are broadly two categories of neural models: compartmental and
point models.
- I have heard biologists call compartmental models explanations.
- Hodgkin-Huxley models (1952) are the most famous. They can be
very accurate and deal with the electrical properties of neurons
and even ion channels.
- I've not done much with them, though I've been trying the
Neuron simulator. They're computationally expensive.
- Point models are much simpler.
- McCullouch Pitts neurons are integrate and fire.
- Hopfield uses these models, and a lot of work makes use of
statistical mechanics for proof; this often assumes well-connected
nets with bi-directional connections. A Hopfield stable state is
akin to a CA.
- There is also a lot of work with Leaky Integrate and Fire neurons.
- We're rather unusual in using fatigue, though not unique. Also, I
think some of the compartmental people are using it.
- Most models also tend to use finer grain time constants.
- Boltzmann machines are also point models.
- These all spike or fire, but there are also rate coded models.
