In article <LVzrd.124203$V41.19112 at attbi_s52>,
"AngleWyrm" <no_spam_anglewyrm at hotmail.com> wrote:
> Yes, with respect to neural nets and artificial intelligence. I have seen
> many such networks, and they involve a set of input neurons, which pass
> signals to one or more layers, and then on to an output set. What I'm
> wondering is if this directedness between neuron layers is an accurate
> model. I have seen pictures of neurons, which look like a complex root
> structure of inputs, and it gets me wondering.
Well, in a system like the visual system, there is a kind of layer
hierarchy in that receptors transmit to ganglion cells, ganglion cells
transmit to thalamic neurons via the optic nerve, and thalamic neurons
transmit to cortex, which then breaks the information into various
streams. However it's far from unidirectional. I'm not sure about the
optic nerve, but at every other stage of the system there are
projections flowing in the opposite direction - for instance Sillito has
estimated that there are more fibres coming BACK from the visual cortex
to the thalamus, than go from thalamus to cortex.
> If axons can be long (as in longer than just adjacent neurons), do some of
> these axons feed signals to neurons that--either directly or indirectly
> through other neurons--supply inputs to that very neuron? More technically,
> is it correct to consider the brain as containing cyclic directed graphs?
Well, yes and no. In many systems there are lesser degrees of
hierarchical organisation, and it is quite common for neurons from one
area to project to widely differing parts of the nervous system, such
that there is more than one path from point A to point B, and it might
involve different numbers of synaptic connections. In this sense it
might fit the definition of a cyclic graph. However the connectedness
of neurons to each other is far more complex than a simple graph - for
instance, a neuron may project to a neighbouring region AND a distant
region using the same axon with multiple terminations. Neurons also
form synapses with themselves, such that their axon contacts one of
their own dendrites.
I think it's safest to assume that a neural network bears little
resemblance to the reality of the nervous system! They are useful tools
with semi-interesting properties, but they don't tell us anything about
the nervous system per se.