[Comp-neuro] From Socrates to Ptolemy
bower at uthscsa.edu
Fri Aug 1 15:12:38 CEST 2008
One more point which just came up in a student presentation at LASCON just now. This student, an engineer who previously had worked on abstract neural network model, has spent the last 3 weeks working the the realistic turtle visual cortex model of the turtle built by Phil Ulinsky over the last many years.
As a neural network modeler would, he gave different types of input to the model to see if the model generated different types of output in response. It did, and in his conclusions he said he was surprised to see this behavior, given there was no "learning rule" in the model.
I asked, and will now ask here to a larger group, what is the evidence that cerebral cortex "learns" in the neural network sense at all. Brain structures represent learning over evolutionary time. Photo receptors don't have to "learn" to detect photons, and the retina doesn't have to "learn" its fundamental structure. Why do we think that cortex has to "learn" most (maybe vastly most) of what it does. We now suspect that the olfactory system already "knows" about the metsbolic structure of the natural world.
Anyway, the point for realistic modeling is that the structure of the brain already includes this evolutionary learning. One can make the case that abstract modelers are in some sense themselves recapitulating phylogony with much more uncertainty, and probably also a less rigorous mechanism for selection, than biology. And in addition, of course, biology sort of had a head start. ;-).
By the way, there are very practical examples of this. I would claim that the more abstract models of cortical oscillations are only now starting to reveal the relationships that were already clear in Matt Wilson's model of olfactory and visual cortex built 25 years ago.
From: G. Bard Ermentrout
To: James Bower
Cc: comp-neuro at neuroinf.org
ReplyTo: bard at math.pitt.edu
Sent: Aug 1, 2008 6:28 AM
Subject: Re: [Comp-neuro] From Socrates to Ptolemy
We've established that there is no "noise" in the nervous system. Now lets
take on the shibboleth of "realistic" models. So, I will ask you all
why a model with 10000 compartments with dozens of active channels, none
of which has been measured (or probably can be with current techniques) is
more realistic than an abstracter model about which one can prove or argue
with some rigor is capable of explaining the underlying phenomena. I think
one can easily go to far in simplifying, but one can also err in the
Sent via BlackBerry by AT&T
More information about the Comp-neuro