Query: Dendritic Networks

Simon R. Schultz simon.schultz at psy.ox.ac.uk
Tue Feb 6 05:55:31 EST 1996


> 
> In article <mike.lowndes-0102961103290001 at mac3.anat.ox.ac.uk>
> mike.lowndes at anat.ox.ac.uk (Mike Lowndes) writes:
> 
> >   You can bet yr bottom pound sterling that anything written about it by a
> >   physicist/mathematition/ modeller is almost bound to turn out WRONG.  The
> >   only way to get it right is to understand the physiology as a first step
> >   and we as neurscientists don't even understand that.
> 

OK, I've been drawn, I have to weigh into the argument here. The reason that
neuroscience for most of its history has concentrated on churning out reams of
details, rather than on understanding, is because *understanding is so much
harder*. (At least if you've got any sort of commitment to being right). If we
wait until every last detail of neurophysiology has been worked out before we
begin trying to understand the system, then I suspect the time when we will
actually understand *how the brain works* is _never_. So what I am saying is:
yes, the physiology is fundamental, something that every brain scientist
should be extremely intimate with, but while the physiological data is still
being produced, effort should be made to bring every scientific technique we
possess to bear upon the problem of analysing that data. And many of those
techniques - statistical, statistical-mechanical, information theoretic etc.,
are techniques which physiologists on the whole simply don't have the
background to master.

There is room for specialisation in neuroscience, and there's plenty of room
for those who specialise in theoretical analysis -- if these people (such as
myself) spent 2/3 of every day running recordings, they simply wouldn't have
time to develop their theoretical skills sufficiently.

On the other hand, the physiology (and anatomy) *is* fundamental. Lets have no
sympathy for those neural modellers who, despite knowing that their models
conflict with the physiology (which is a different thing to just not including
all the details for the sake of simplicity), persist with them. We all know
who they are.

On the other hand, there isn't always an advantage to detailed (e.g.
compartmental) modelling either. Firstly, this sort of level of modelling, in
all except the most tightly constrained circumstances, gets in the way of
understanding what the system is doing - for that more simple approaches
(possibly in combination with the detailed modelling) are needed. Secondly,
the increase in degrees of freedom of these models means that these models
aren't necessarily any more biologically realistic than slightly simpler
models in any case, since the extra parameters cannot be measured in most
cases. They do have a role, but it shouldn't be taken too far. 

-- 
-----------------------------------------------------------------------
Simon R. Schultz                       |  Research Interests:
Department of Experimental Psychology  |    Computational Neuroscience
(Also Corpus Christi College)          |    Hippocampus/Memory
University of Oxford                   |    Integrated Circuit Design
South Parks Rd., Oxford OX1 3UD, U.K. ---------------------------------
              http://www.mrc-bbc.ox.ac.uk/~schultz/
-----------------------------------------------------------------------



More information about the Neur-sci mailing list