"Andy Glew" <aglew at students.wisc.edu> writes:
> I keep hearing this about analog neurons,
> but my understanding is that biological neurons
> are pulse systems - discrete signal, possibly continuous
> time. Probably emulatable with high clock frequencies.
Information transfer from neuron to neuron over more than a few mircometers
is "binary" - action potentials are on or off. Integration on the dendritic
tree, which leads to the "decision" to send an action potential (or not), is
definitely analog and highly nonlinear. In all likelihood, feedback of the
neuron's "decision" to fire is sent back up the dendritic tree more often
than not.
Of course, there are also electrically-coupled neurons doing analog computing.
These seem to turn up more and more, for instance in the retina.
> More interesting is how stateful a neuron is - apart from reconfiguration,
> how much of an integration function and delat there is.
Quite so. Every spiny synapse in all likelihood is its own compartment with
its own short- and medium-term state. And don't forget protein sysnthesis,
regulation and transport. Makes the mind boggle...
Jan
