Neurons
- A big part of my research is working with computataional models
of neurons.
- Point Models
- Simple models are called point models, modelling a neuron
as an equation.
- There are a lot of point models, but a simple one is the
integrate and fire model.
- It's called the McCulloch Pitts model (1943) but Lapicque did
a variant in 1907.
- If the weighted sum of presynaptic firing neurons is greater than
the threshold, fire.
- It's a (really quite) simple equation of neural behaviour.
- It's efficient (to simulate) and pretty accurate.
- Compartmental Models
- Point neurons entirely ignore that a neuron is a cell with
structure that is important to its behaviour.
- A more accurate model is a Hodgkin-Huxley model (1952 which
was the basis of their Nobel prize).
- It's also called a conductance based model and it models a
neuron as a series of compartments with a flow of current
between the compartments.
- Neurons behave differently from each other and one big difference
is because of different size and shape.
- Some neurobiologists call these models descriptions (while point
models are just models).
- There are other models that deal with neurotransmitters, and
the change of neurons over longer times (days) to deal with
learning.
- Improving Models
- There is work on developing better models of single neurons.
- I'm aware of work simulating a day's behaviour of a neuron
using several years of computation (on a super computer).
- There is ongoing work on developing bettter complex models.
- There is ongoing work on developing more computationally efficient
models.
- There is work in developing hardware for emulating neurons.
- So which is best?
- It all depends on what you want to do with a neural model.
- If you're trying to model the behaviour of a lot of neurons (e.g. because
you want to model cognition like me) you want an efficient model.
- If you're trying to figure out how synaptic transmission changes, you
want a more accurate model.
- If you're looking at disease, like Alzheimer's or cancer, you want an
entirely different sort of model.