[Comp-neuro] Development, Evolution and Learning (was) Discussion - Kuhn - and brief comments

Asim Roy ASIM.ROY at asu.edu
Thu Sep 4 22:02:17 CEST 2008

Perhaps Eugene can respond to some of the questions here. I am sure he understands and can relate to some of the ideas in artificial neural networks. And we of course learn all the time from new discoveries in biology and neuroscience.<?xml:namespace prefix = o ns = "urn:schemas-microsoft-com:office:office" />


I believe the sense of the argument here is that learning is simply some "fine tuning" of some existing structures. Adult neurogenesis shows that new networks can be created and put into operation. And that's not some simple "fine tuning." And I believe there is plenty of evidence to show that there is remodeling and reconfiguration in the brain (new connections, synaptic adjustments). Perhaps someone can help out with some references. I don't have anything handy.





-----Original Message-----
From: comp-neuro-bounces at neuroinf.org [mailto:comp-neuro-bounces at neuroinf.org]On Behalf Of Mario Negrello
Sent: Thursday, September 04, 2008 2:17 AM
To: CompNeuro List
Subject: [Comp-neuro] Development,Evolution and Learning (was) Discussion - Kuhn - and brief comments

Asim said: 
On the "blank slate" idea in ANN, just look at the phenomenon of adult neurogenesis. Our adult brains generate new cells in the thousands on a daily basis and they are part of the "blank slate" because they don't come with ready-made connections or anything. The process that you refer to as "development probably recapitulating evolutionary history" is called "learning" in ANN. It's that process that constructs networks out of these new cells and makes them operational.

I guess that this varies with your level of analysis. Developmental recapitulation of evolutionary history is about major brain development, hubs and tracts and gross architecture, those features that are morphologically conspicuous in the brain. Learning is different, it's fine tuning, when an already functional brain develops further functionality by interacting with a complex environment. Moreover, as was pointed out by Cottrell, the new neurons you mention do not sprout anywhere, but presumably, noly where there is a reason or a function for them to do. One of the stories i know about new neurons being generated relates to communication behavior in fish. Fish that produce aggressive calls, grow new cells to make their calls more potent. This was solidly shown in:

Dunlap, K.D., et al 2006. Social interaction and cortisol treatment increase cell addition and radial glia fiber density in the diencephalic periventricular zone of adult electric fish, Apteronotus leptorhynchus, Hormones & Behavior 50:10-17.

So, between cell addition and brain development there is a considerable distance. I do not suppose that by explaining mechanisms of learning, we shall also understand the functional entailments of particular architectures. 

For instance, Eugene Izhikevich in a talk in 'Dynamical Systems in Neuroscience' at the NYU in June 2008 (if memory serves me right), showed the last output of his mega brain model. The model has the gross architecture of a human brain as seen through DTI (diffusion tensor imaging), and the microarchitecture derived from rats and cats. What surprised me in terms of the results, was that he claimed that frequencies measured from the model resembled those known from their correspodent areas of the brain. Alpha in V1, Theta in hippocampus, and others.
I take that to mean that many functional capabilities of the brain will be effectively only learneable if the proper architecture is there. Else, if it were only learning, there would be little reason for a cat to be unable to learn the things a human does. 

Surely, one can draw analogies from evolution, through development, until learning.  One can amalgamate all and say that they are all 'metastable adaptations', or something. But so far, the clear divisions that appear between the study of each, are fully justified: we need theories for each. For all, later. Learning needs its story as much as development. 




-----Original Message-----
From: comp-neuro-bounces at neuroinf.org [ mailto:comp-neuro-bounces at neuroinf.org]On Behalf Of james bower
Sent: Wednesday, August 27, 2008 9:05 AM
To: CompNeuro List
Subject: [Comp-neuro] Discussion - Kuhn - and brief comments

A few brief comments -- given that school has started, and I now have to prepare to "influence" the latest next generation.

First a BIG POINT:

Klaus Stiefel:  pre-paradigmatic " What he meant by that is a disagreement about the basic explainanda"

No, Kuhn was focused on process to quote:  "the early developmental stages of most sciences (are) characterized by continual competition between a number of distinct views of nature, each partially derived from, and all roughly compatible with, the dictates of scientific observation and method,  (In pre-paradigmatic science) what differentiates these various schools (is) not one or another failure of method - they are all "scientific" - but what we shall come to call their incommensurable ways of seeing the world and practicing science within it"  (pg 4 The Structure of Scientific Revolution.)   If this discussion over the last two months doesn't make it clear that, as a field, we currently have "incommensurate ways of seeing the world and practicing science within it"  I don't know what does.  Kuhn goes on to say:  "Men (sic) whose research is based on shared paradigms are committed to the same rules and standards for scientific practice."  Again, this discussion makes it pretty clear to me that we have not yet reached that point. In fact,  (and I would say reflecting this fact) the kind of discussion we have been having here seldom ever happens as we are content (and being pre-paradigmatic can get away with) agreeing to not discuss what we don't agree on, another characteristic of pre-paradigmatic science -- and the reason I don't mind starting these discussions.

Bryan Bishop:  "There's a few too many layers of folk psychology here,"

I agree, and have characterized all of biology as fundamentally folkloric in nature -- based on story telling, with few real definitions of anything.  Something comp bio (neuro) will, I hope, eventually fix.

And on to the approaches we don't agree on:

Bryan Bishop:  "I bet it  becomes clear that trying to do "natural language processing" from statistical inferences doesn't get us as much hard science as the brain could provide."

Of course, I agree -- and also agree that this field continues to be distorted by what is essential snake oil we sell to the Department of Defense about how studying the brain will help win wars.  This rather self serving commitment to "neuro-morphic engineering" as it is now called, has been distorting our science for a while.  In the last 8 years even more dramatically.  

Mario Negrello:  "I'd say instead that some approaches gather more acolytes, and then overflow others in sheer voluminous quantity,"

Unfortunately, as just noted, very often related to who you are selling the science to (funders) and also, unfortunately, how 'easy' the methods are and therefore how many can jump on the bandwagon without much preparation (or even knowledge of the brain in this case)   -- 20 years studying the realistic model of the Purkinje cell and counting.  

Bard Ermentrout:  " I suspect that it would be too hard to adjust parameters for realistic models "

Is hard bad?  Or is the brain, in fact, hard?

Igor Carron: "The groundwork of theory as you put it has, in nuclear technology, always been a way to acquire and use experimental findings." 

And this, in fact, is the value of theory -- not to capture 'truth' as many in comp neuro seem be believe, but to organize experimental studies -- the more the theory is removed from the actual structure of the brain - the more it exists by itself, disconnected from the ability to improve, or more importantly to refute it.

Asim Roy:  "a "blank slate" simply implies a network whose connection weights and other parameters have not been set yet -"

And, no such thing in biology, where development doesn't end and learning begins - it is continuous, and development probably recapitulating evolutionary history.  Another reason why the mainstream ANN models make no sense.

Asim Roy: "Is there a way in computational neuroscience to verify any of these theories of learning? "

Wrong question.

Axel Hutt:  " can (neuro)biology really treat a population of some thousand elements ? "

We will need to figure out how - numerous groups are working on it.

Anibalmastobiza:  "cerebellum, usually considered as a center for motor processing and coordination just  as it was for the basal ganglia that now we know that is also involve in cognition"

While I appreciate the support, I have another question for cognitive neuroscientists, how come anything that lights up in a brain scan becomes a "cognitive center"  seems weird to me.  



Dr. James M. Bower Ph.D.

Professor of Computational Neuroscience

Research Imaging Center
University of Texas Health Science Center - 
-  San Antonio
8403 Floyd Curl Drive
San Antonio Texas  78284-6240

Main Number:  210- 567-8100
Fax: 210 567-8152
Mobile:  210-382-0553 

The contents of this email and any attachments to it may be privileged or
contain privileged and confidential information. This information is only
for the viewing or use of the intended recipient. If you have received this
e-mail in error or are not the intended recipient, you are hereby notified
that any disclosure, copying, distribution or use of, or the taking of any
action in reliance upon, any of the information contained in this e-mail, or
any of the attachments to this e-mail, is strictly prohibited and that this
e-mail and all of the attachments to this e-mail, if any, must be
immediately returned to the sender or destroyed and, in either case, this
e-mail and all attachments to this e-mail must be immediately deleted from
your computer without making any copies hereof and any and all hard copies
made must be destroyed. If you have received this e-mail in error, please
notify the sender by e-mail immediately.

Comp-neuro mailing list
Comp-neuro at neuroinf.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.neuroinf.org/pipermail/comp-neuro/attachments/20080904/f426ee5d/attachment.html

More information about the Comp-neuro mailing list