plausibility

[James M. Bower]

In response to my recent post, I received the following note:


>> actually designed to demonstrate.
>                          ^
> You surely meant to say test.
>


In fact I meant "demonstrate", and this is a very important distinction and
issue in brain-like modeling.  It is my view that the majority of the
models generated in this field to date (especially those of the NN type)
are actually demonstration type models, and not in any real sense "tests".
In order to be a test, there must be some mechanism for formally evaluating
the plausibility of a particular model, given the available neurobiological
data.  We have recently published a paper suggesting one (some would say
the only) formal approach to this problem:

Baldi, P.,  Vanier, M.C., and Bower, J.M.  (1998)  On the use of Bayesian
methods for evaluating compartment neural models.  J. Computational
Neurosci. 5: 285-314.

However, at present there are no accepted standards for such an evaluation
(in fact there is almost no discussion of this issue).  Instead, far too
much modeling involves twiddling the right knobs to get the functional
results you want.  Those few experimentalists interested in modeling
usually evaluate a models plausibility based mostly on intuition.

It is for this reason that the question of prior functional assumptions is
so important, and why I continue to try to draw a strong distinction
between modeling based first on anatomy and physiology and efforts intended
to demonstrate the plausibility of a particular preconceived idea (c.f.
Bower, J.M. (1995)  Reverse engineering the nervous system: an in vito, in
vitro, and in computo  approach to understanding the mammalian olfactory
system.  In: An Introduction to Neural and Electronic Networks, Second
Edition. S. Zornetzer, J. Davis, and C. Lau, editors.  Academic Press.  pp.
3-28.).  If the functional idea truly comes about as a result of the
modeling, and not vice versa, then it is more likely (although still far
from certain) that the revealed mechanisms have something to do with the
real brain.  Of course, this is the same reason that some modelers try to
blur this distinction.


Jim Bower

================ Message 2 of 4 ================