The questions you ask actually require quite speculative answers, however,
I will throw in one or two words:
> What are the processes involved in the early stages of brain development,
> which are essentially the hard-wiring of nerve connections between
> different specialised areas of the brain (e.g. eye nerves to vision
There are many different processes. The one that is probably interesting to
you is that in general it is necessary for sensory input to come in
for a sensory system to develop normally. The first impressive data
collected on this was by Hubel & Wiesel, where they showed that if
they blindfolded a cat it would not develop normal neural connections.
People later (especially Merzenich) showed that there was also remarkably
plasticity in adulthood. For example, he showed that if you fused two
fingers together their cortical representations in sensory cortex would
fuse (many other interesting experiments in this line). There is probably
a lot of applicability of neural network kind of stuff to this, since there
seems to be a lot of top-down regulation of how the neural input is organized.
>> How is this achieved? More importantly, how did this pre-determined
> structure evolve? Is it genetic, or something else?
Well, noone really thinks all of the connections in the brain are mapped
out genetically, certainly, because there wouldn't anywhere near enough
genes. However, strategies for neural development must be coded
genetically. These are not invariant, as damage to a system often
results in an alternative generation rather than dysfunction. For example,
patients whose corpus callosum is split in adulthood show the classic
split-brain syndrome. On the other hand, children born without a
corpus callosum (callosal agenesis) show relatively normal interhemispheric
transfer. There are many other examples of this. So the direction of
development depends on variable genetic strategies, but requires sensory
experience (and probably learning), at least for many systems.
>> What is the mechanism involved in determining nereve connections (in
> learning and memory, if I'm not being too naive about our current
> knowledge of how the brain works)? What chemical or electrical signal
> encourages a particular neurone to want to send feelers along to
> g=connecto up with another? What is it that makes one nerve more
> 'interesting' than another? Is there such a process, or is it all part of
> the hard-wiring phase?
Well, the interesting one in mammals is LTP. You should pick up an textbook
on integrating LTP into neural network theory, but it is too detailed to
cover here. Try reading Computational Neuroscience, Churchland & Sejnowski.
There are many plastic synaptic processes in the brain, some of which
have to do with learning, and hard-wiring views of the brian, although
once popular, are now quite unpopular (and wrong).
>> I'm interested because my guts tell me that dynamic neural networks (DNNs)
> are potentially more powerful than the static, forward flow kind that is
> currently in favour. A hunch tells me that the 'learning' and modification
> of such a network of DNNs should be self-modifying rather than having
> modifications imposed upon them by some outside means, as happens in
> Neural Nets at present. I'm looking for how the structures evolve in the
> first place and how they modify themselves (chemical feedback?) during
> maintenance. My ultimate aim is to build a bloody good model of learning
> and symbol fixing using a fairly low level representation mechanism (an
> ambitious objective I know). Thank you for your time.
Good luck, you are not alone. Certainly half of southern california has
been taking over by you neural net types!
> Yours Mike Reddy
> Email: mreddy at comp.glamorgan.ac.uk CU-Seeme: 184.108.40.206 (On Request)
> Web: http://www.comp.glam.ac.uk/pages/staff/mreddy/mreddy.html> Snail: J228, Dept. of Computer Studies, University of Glamorgan,
Pontypridd, Mid Glamorgan. CF37 1DL Wales, UK. +44 1443 482 240 Fax: +44
1443 482 715
STEPHAN ANAGNOSTARAS UCLA BEHAVIORAL NEUROSCIENCE
STEPHAN at PSYCH.UCLA.EDU