[Comp-neuro] The sniffing brain - and free will
james bower
bower at uthscsa.edu
Fri Aug 15 18:33:59 CEST 2008
Asim,
Glad you came out in the open :-)
> The question here is related to the first one: Is there a way in
> computational neuroscience to verify any of these theories of
> learning?
This, of course, is exactly the 'top down' kind of question that
continues to worry me. Of course, there is an enormous amount of work
done in learning mechanisms in neurobiology. It is a huge literature
at all levels of scale. There are also abundant examples of the
influence of top down theories on brain science -- the one I
personally have to deal with being the Marr/Albus theory of cerebellar
learning -- which has produced an entire industry of neurobiologists
intent on proving the idea to be correct -- even in the face of
abundant evidence (even their own) to the contrary).
However, I have been deeply concerned for a long time that our
emphasis on the importance of learning really derives from our belief
(hope) in free will. The Stealth Duelism of many cognitive brain
models, I suspect, is similarly related to the deep held conviction
that humans with our big brains, somehow operate as free entities,
organizing our cognitive structures based on unique solutions to our
own pattern of inputs. See the posting today on variations in
cognitive structures. This, of course, has been a particularly
prevalent view in the largely protestant United States. (Ah, how I
wish the Portuguese rather than a bunch of English / German religious
zealots had landed at Plymouth Rock).
For those of you still left, who haven't rolled your eyes and just
sent an angry letter to the moderator asking "when is enough enough",
if you are involved in studying something as near and dear to hominids
as the brain - I am afraid that one has to seek and consider the
influence of that brain's own predispositions on its study of itself
(as has been pointed out by philosophers for thousands of years).
But I don't raise this issue simply as a sophomoric exercise to invoke
nostalgic feelings about college dorm rooms -- Earlier I asked what I
consider to be a very serious and practical question -- what is the
evidence that most of the behavior we ascribe to learning actually is
due to the kind of learning mechanisms being studied by
neurobiologists and dear to the hearts of the neural network and
abstract modeling community? Sure, we can force starved or thirsty
monkeys to learn weird things (and do weird things), but how much of
the real behavior of real brains in the real world has anything to do
with the "learning" we hold near and dear to our hearts? Neural
network guys want stuff to learn, because it makes engineering easier
(or does it??), but what do we really 'learn". Evidence in the US at
the moment, is not very much.
Cognitive psychologists know about this well - human phobias involve
snakes and spiders, not electric outlets, despite the fact that
electrical outlets are more numerous and more dangerous. Although
still described as 'learning' even in the "associative learning
domain" it is much easier to train humans to associate pictures of
snakes and spiders with aversive stimuli than more dangerous real
world objects like electrical outlets. As I pointed out with respect
to the recent NSF report cited on this list -- the question of what we
already "know" and how it governs our behavior, or even the extent to
which what looks like 'learning' might actually be a different fixed
read out, with changing context, is seldom considered by computational
neurobiologists, or the neural network community. We seem fixated on
the idea that we are "learning machines" and largely lumps of clay
fashioned by experience (especially again in the US -- Conrad Lorenz
didn't think so at all). This of course, is simply silly. Please
note that this question is actually deeply connected to the abstract
vs realistic modeling issue -- if, in fact, the structure of the
nervous system is highly patterned based on a long evolutionary
history, THEN ignoring that structure when building models may
profoundly miss the point.
To be even more specific, I now suspect that the olfactory system may
actually have built into it, at birth, a great deal of knowledge about
the detailed structure of metabolic systems in the real world --
While the olfactory cortex has been considered for many years
(including by me), as being as close to a pure associative learning
network as can be found in the brain, in fact, I now suspect that the
appearance of an unstructured highly interconnected pattern of
neuronal connectivity, may actually be disguising a highly ordered set
of connections reflecting prior knowledge about the metabolic
structure of the real world. This fundamental change in my own
thinking was driven by results of an olfactory cortical model that is
probably the most complex realistic model that has so far been
constructed. Unfortunately, the work has only been published in the
form of a thesis (M. Vanier, Realistic computer modeling of the
mammalian olfactory cortex. http://nsdl.org/resource/2200/20080620191954527T
). Experimental evidence that the olfactory system might "know" a
great deal about bio-metabolism is also only published in thesis form -
Ruiter, Christine, “The Biological Sense of Smell: Olfactory Search
Behavior and a Metabolic View for Olfactory Perception”, Ph.D.thesis,
California Institute of Technology, Pasadena, CA,. But here is a link
to some subsequent work http://www.inb.uni-luebeck.de/forschung/eops/INS2002.pdf
that you might find interesting.
The point being, that where the number of neurons are small
(invertebrate systems), development produces highly order and
reproducible networks. Most mammalian neurobiologists have asserted
that brains with large numbers of neurons have adopted a different
strategy that is more flexible and plastic (read 'free will"). I am
not so sure. The complexity of the mammalian brain makes it very hard
to ask the question.
One last point about free wheeling cognitive possibilities - I don't
believe it for a minute. In fact, I suspect that our cognitive
structure is fundamentally linked to the computational problem faced
by the olfactory system, and the computational solution it reached to
solve that problem. In other words in some computational / cognitive
sense 'we sniff the world' whether we use visual data, auditory data,
somatosensory data, or olfactory data. For implications see
( Fontanini, A. and Bower, J.M. (2006) Slow-waves in the olfactory
system: an olfactory prospective on cortical rhythms. Trends in
Neuroscience. 29: 429-437 ). Thus, in fact, I don't believe at all
that we have huge cognitive freedom, I think we are fundamentally
constrained by how the olfactory system "thinks". Its just that our
inordinate focus on the visual system (visual primates with large
brains) has lead us seriously astray in thinking about how brains work.
Writing a book on this -- thanks for helping.
Jim Bower
On Aug 14, 2008, at 3:55 AM, Asim Roy wrote:
> Hi All,
>
> I read with interest this ongoing debate in the computational
> neuroscience community. Not being a computational neuroscientist
> myself, I was a little hesitant to wade into this debate. However, I
> thought I would raise some issues of deep interest to me and try to
> get some feedback from the community.
>
> My hunch is that one part of computational neuroscience is about
> discovering the existing wiring and operating mechanisms of certain
> parts of the brain that generally come predefined or prewired to us,
> like parts of the vision system. Some modules that may not come
> predefined or prewired are like the ones that a biologist has to
> create in his/her brain to learn a bit of mathematics in graduate
> school. That stuff is new for the brain and it's hard to learn. Wish
> it came prewired.
>
> So I hope that there is another side of computational neuroscience
> that looks at learning mechanisms within the brain. That's the side
> that is of great interest to many of us who work on learning
> algorithms. I was wondering if there are any new insights or
> theories in computational neuroscience on how the brain "learns."
> Here are some issues that I would love to get some feedback on:
>
> 1. The artificial neural network community still believes that
> learning in the brain is real-time, almost instantaneous. It's real-
> time in the sense of Hebbian-style learning. And I believe
> computational neuroscience predominantly uses Hebbian-style models
> of learning. I personally doubt that the brain learns in real-time.
> There is plenty of evidence in experimental psychology to refute the
> real-time learning (Hebbian-style synaptic modification) claim. And
> there is also enough recent evidence in cognitive neuroscience too
> to refute that claim, although one has to carefully read between the
> lines the conclusions of these papers. (In one such case, I
> suggested a different interpretation to the results and the authors
> agreed with it.) One can also logically argue that real-time
> instantaneous learning amounts to "magic" since no system,
> biological or otherwise, can set up a network and start learning in
> it without knowing anything about the problem before the start of
> learning.
>
> My question is, is computational neuroscience still a firm believer
> in Hebbian-style real-time learning or have researchers looked at
> other forms of learning, like memory-based learning that is not real-
> time?
>
> 2. It appears that the brain has the capacity to design networks
> when a new skill has to be learnt. Are there any studies/insights in
> computational neuroscience on how this design process works?
>
> 3. In machine learning and neural networks, there are two extremes
> sides to designing of algorithms. At one end are back-propagation
> type algorithms where neurons in the network use local learning laws
> to learn. At the other end are Support Vector Machines (SVM) type
> algorithms, which were mentioned in that NSF report, which bring in
> heavy computational machinery (e.g. quadratic programming) to both
> design and train neural networks. SVMs don't use local learning
> laws. I don't believe we have SVM-style mechanisms in our brains;
> it's just too complicated. So SVM algorithms are unrealistic for the
> brain, although they are widely used to solve learning problems for
> the machine learning community. But Hebbian-style or back-
> propagation-style real-time learning also has problems and that's
> not just with evidence from cognitive neuroscience and experimental
> psychology, but logical ones too.
>
> The question here is related to the first one: Is there a way in
> computational neuroscience to verify any of these theories of
> learning?
>
> Hope I am not asking stupid questions. Would love to get some
> thoughts and feedback. And any references would help.
>
> Best wishes,
> Asim Roy
> Arizona State University
>
> _______________________________________________
> Comp-neuro mailing list
> Comp-neuro at neuroinf.org
> http://www.neuroinf.org/mailman/listinfo/comp-neuro
==================================
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
CONFIDENTIAL NOTICE:
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.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.neuroinf.org/pipermail/comp-neuro/attachments/20080815/a625e57a/attachment-0001.html
More information about the Comp-neuro
mailing list