Copy Link
Add to Bookmark
Report
Neuron Digest Volume 04 Number 34
Neuron Digest Thursday, 22 Dec 1988 Volume 4 : Issue 34
Today's Topics:
Reinforcement Schemes for Learning Automata
Haverford job
Re: Some biological questions
Re: some biological questions (Neuron Digest V4 #33)
Re: Some biological questions
SBIR on parallel processing
Question for you or your viewers
B.P. nets as associative memory nets.
Send submissions, questions, address maintenance and requests for old issues to
"neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request"
------------------------------------------------------------
Subject: Reinforcement Schemes for Learning Automata
From: DMEREDI3@UA1VM.BITNET (Don Meredith)
Organization: The Internet
Date: 08 Dec 88 19:26:07 +0000
I am in desperate need of some literature on reinforcement schemes
for learning automata and weight changing algorithms for neural
networks. You can send any information to DMEREDI3@UA1VM or to the
following address:
Don Meredith
1049 Taylorwood Circle
Tuscaloosa, AL 35405
------------------------------
Date: Mon, 12 Dec 88 10:00 EST
From: <J_SCHULL%HVRFORD.BITNET@CUNYVM.CUNY.EDU>
Subject: Haverford faculty position
FACULTY POSITION IN BIOPHYSICS AT HAVERFORD COLLEGE
The Department of Physics invites applications for a special
assistant or associate professorship in biophysics for a period
of up to 4 years. Applicants must have a Ph.D. and be strongly
motivated toward creative research and undergraduate teaching.
The successful candidate will assist in developing a new
interdisciplinary biophysics program. Substantial startup funds
for research are available. The position is part of a major
program funded by the Howard Hughes Medical Institute to enhance
the biological sciences through interaction between the various
science departments.
Haverford is a liberal arts college with an international
reputation for strong research and instruction in the physical
and life sciences. We have active collaborations in the sciences
with the University of Pennsylvania, Bryn Mawr College, and
Swarthmore, among others. Please send curriculum vitae, list of
publications, a statement of research interests, and at least
three letters of recommendation by 16 January 1989 to: Professor
David Pine, Department of Physics, Haverford College, Haverford,
PA 19041. An Equal Opportunity/Affirmative Action Employer.
------------------------------
Subject: Re: Some biological questions
From: "Neuron-Digest Moderator Peter Marvit" <neuron-request@hplms2>
Date: Tue, 13 Dec 88 17:05:51 -0800
Unfortunately, the answers to most of these questions are not as simple as
the questions themselves. The computer, however astonishing even to
practitioners of the the art, pales in subtle complexity compared to dynamic
living systems. Our knowledge of the nervous system is patchy at best; only
the largest neurons have actually been recorded from and only an estimated
100 mm^2 of the 1m^2 cortex has been carefully and systematically mapped.
It would not be an exaggeration to say that Neuroscience is pre-Newtonian.
Given those caveats, I'll attempt a reasonable first-order. My knowledge is
not deep and so detailed supplements with citations are welcome. There is a
vast literature extant, and I will commend two books to start with.
"Principles of Neural Science," Kandel & Schwartz, eds. (Elsevier 1985) is
considered by some to be the "Bible"; while not always the clearest writing,
this survey covers a vast amount of material in its 979 pages and provides
an excellent overview of all parts of Neuroscience. "From Neuron to Brain",
by Kuffler, Nicholls, & Martin (Sinauer, 1984) takes a more cellular
approach, with special concentration on the visual system as paradigmatic of
the rest of the nervous system; generally quite well written, it offers an
excellent introduction to the workings of the neurons themselves. Both
books give extensive "Suggested Readings" and bibliographies. Most of the
following comes from these two books.
> How diverse are the neurons in a small system of neurons (or in
> selected regions of the brain)?
Current estimates of morphologically distinct neurons number near 10,000;
that is, there are about 10,000 different types of neurons. In addition,
there are some dozen types of glial cells (relatively recently discovered),
even though in population they outnumber neuron 50:1! The diversity of
neurons depends on the region and its extent. The cerebellum, one of the
best studied structures, is an extremely regular arrangement of only a few
types of neurons (Purkinje cells, basket neurons, stellate cells, Golgi
cells, granule cells); Purkinje cells provide the sole output of the
cerebellum (entirely inhibitory), but each cell has about 150,000 dendrites
(external contacts) and receives input from 200,000 contacts!
The somatogastric system of the Pacific spiny lobster is one of the best
understood of all invertebrate neuronal systems. Researchers in San Diego
(amongst others) have been working working to describe its mechanisms. The
gastric mill has 12 neurons of 7 types which appears to have rythmic
properties not predictable from the individual cells; the pyloric muscles
are controlled by 14 neurons of 6 different types in a complex web which has
only recently been characterized as to its functional interconnections.
The short answer is: incredibly complex!
> Can somebody give me a general idea of the complexity of the
> chemical reactions that occur in the cell body? (A vague
> question I know, but I'm just trying to get an idea how much
> is going on in there).
The general mechanisms of signal generation and transfer down a neuron's axon
are apparently straight forward -- dealing with voltage sensitive ion
channels and simple physics. The subtleties of neuronal cell metabolism and
ionic pumps are as complex as any other cell, with the added feature of
maintaining an electrical potential. A useful first reference is "Molecular
Biology of the Cell", Alberts et al, eds. (Garland Publishing, 1983)
although many of the finer details are rapidly becoming out of date.
> Approximately, how many chemicals/ions have been found in and
> around a neuron?
Again, the basic constituants of signal transfer involve (nominally) 4 ions:
K+, Cl-, Na+, and Ca++. However, over 60 transmitters have been
identified... often several acting within a single neuron. Further, neurons
are subject to the vast chemical arsenal of the human body -- hormones,
peptides, enzymes, amines, proteins, etc. I would guess that one could tally
well over 100,000 chemicals and ions which react with cortex neurons.
> Is it true that the basic structure of the brain is determined
> when you are born?
Actually, much is before you are born. The physical structures of the brain
is remarkably consistent from person to person (as far as we know), and much
of the basic anatomical structure exists before birth. For example, in 95%
of people, one hemisphere (the "dominant" or language or left side) is
physically larger than the other and this difference can be seen in 6 month
old fetuses. However, the brain is enormously plastic in terms of function;
even if the entire dominant hemisphere is removed from an infant, it will
develop relatively normally (complete with full language!) -- contrary to
the extreme deficits one might expect.
> How does the shape of the neuron affect its "computation"?
Really this is unknown. The number of input and output contacts, the size
of the cell body, whether the axon is myelenated, whether is has an axon at
all, all are factors in a cell's response. One can make some
generalizations, based on a neuron's arborization (see different retinal
ganglia for examples), but details are still hazy. The fact is, though, that
we don't understand what a "computation" really is at the neuronal level --
at least near the cortex. In short answer, there must be some reason to
have 10,000 distinct shapes.
> Finally, has anyone determined what role, if any, DNA might
> play in the processing performed by a neuron?
Again, many clues exist, though mostly for early development. I don't have
any citations, but Dr. Jeff Winer at Berkeley agrees with me that DNA
probably has little to do with the moment by moment lives of mature neurons.
I think the "Selfish Gene" concept of DNA guiding all levels of biological
activity from molecular to behavioral must be taken metaphorically; except
in certain narrowly prescribed instances, the genes themselves cease to
exert direct influence past the initial stages of cell differentiation.
> If anyone is interested, this questions were raised during a
> reading of the first two chapters of James S. Albus' "Brains,
> Behavior, and Robotics".
I'm intrigued. I hope others have additional insights.
------------------------------
Subject: Re: some biological questions (Neuron Digest V4 #33)
From: Paul Davis <davis%mauve.sdr.slb.com@RELAY.CS.NET>
Date: Wed, 14 Dec 88 04:48:00 -0400
> Can somebody give me a general idea of the complexity of the chemical
> reactions that occur in the cell body? (A vague question I know, but
> I'm just trying to get an idea how much is going on in there).
Hmmm, depends on what you mean by complexity. A neuron is a cell pretty much
like most others, and as such has an enormously complicated panoply of
(bio)chemical systems in action all the time (gene expression, protein
synthesis, energy metabolism, ionic regulation (intimately related to their
primary function) etc.)
In terms of the reaction systems directly related to neuronal function
(though its debatable whether one can distinguish between those directly
related and those that are not), I would have said that they are not really
`very complex' in the sense of say, a comparison with gene regulation or
metabolism, but they are still enormously complex in comparison to inorganic
chemistry or even simple enzyme-catalysed reactions. If you're thinking of
trying to model them, my recommendation (only as an ex-biochem graduate
student now in CS) would be to forget the details and concentrate on the
feeedback systems implicit in their operation. Alberts et al. "The Molecular
Biology of the Cell" (2nd edition) is a highly recommended starting point
for grasping the intricacies of this stuff - almost bedtime reading.
Paul
Schlumberger Cambridge Research
Cambridge, England
internet: davis%mauve@sdr.slb.com
------------------------------
Subject: Re: Some biological questions
From: reinke%uicslsaj.csl.uiuc.edu@uxc.cso.uiuc.edu (Robert Reinke)
Date: Thu, 15 Dec 88 08:09:18 -0600
[[Editor's Note: This came in after I wrote my response. He repeats some of
my reply, but had additional and valuable insights. -PM]]
In response to Chip Roberson's posting:
These are interesting questions, but hard to answer simply. Biological
systems are *not* simple, and easy generalizations are almost always wrong
in some cases. I have tried to give succint answers to your questions, but
have not included references. The level of detail you're looking for can
best be found in an introductory neurobiology text. If you are really
interested in these questions, I suggest you read such a text (or better,
take an intro course).
> How diverse are the neurons in a small system of neurons (or in
> selected regions of the brain)?
There are certainly hundreds, if not thousands, of anatomically
distinct types of neurons in the vertebrate nervous system.
Some regions of the nervous system, however, do use only a few types.
The classic example (and one reason it is much studied) is the cerebellum,
which contains only five types, connected in a very ordered way. Another
example is the retina, which also contains five distinct cell types.
> Can somebody give me a general idea of the complexity of the chemical
> reactions that occur in the cell body? (A vague question I know, but
> I'm just trying to get an idea how much is going on in there).
I assume you are interested only in the electrical reactions. The
basic mechanisms have to do with ionic channels in the cell membrane.
A grossly simplified outline is that communication between neurons
(usually) occurs through release of neurotransmitters (there are quite
a few different ones known) that interact with membrane proteins, which
in turn activate or inactivate ionic channels in the membrane. Changing
the flow of ions in and out of the cell changes the electrical
potential across the membrane. This potential difference may (at other
places) activate voltage sensitive channels, allowing propogation of
the potential down the cell.
This is really a tremendous simplification. There are a wide variety
of different mechanisms known; the only constant is that electrical
characteristics of neurons are based on the activation and inactivation
of ionic channels in the membrane. There also seem to be more complex
effects involving chemical intermediaries inside the cell that
*permanently* change the channel characteristics.
> Approximately, how many chemicals/ions have been found in and around a
> neuron?
Again, generalizations are tough here. Four ions seem in most cases
to be the ones involved in electrical action: K+, Na+, Cl-, Ca++.
But, there are also the neurotransmitters (common: acetylcholine,
gamma-aminobutyric acid (an amino acid), serotonin,
substance P (a peptide), epinephrin,...) and any other messengers
(e.g., hormones) that can interact with membrane proteins and therefore
affect the electrical properties of the cell.
> Is it true that the basic structure of the brain is determined when
> you are born?
Yes and no. Yes, neurons during development form up in "fixed"
patterns. For example, it has recently been shown (see articles
by Goodman in the Nov. 1988 issue of Science) that neurons form
bundles (fasciculate) based on proteins on the surface of other
neurons. It is also known that in some cases developing axons
seek out "guidepost cells" to find their way to the appropriate
place. On the other hand, it is well known that formation of
connections can be affected by the environment. The classic
example is connections between the eye and the visual cortex
in the cat: it has been shown that if one eye of a kitten is sutured
shut during a critical period (a week approx. 3 weeks after birth),
the sutured eye will be effectively blind after the sutures are
removed, even though the retina is fine. The problem seems to
lie in the visual cortex, where it has been shown that the sutured
eye lacks the normal connections.
> How does the shape of the neuron affect its "computation"?
I'm not even going to try to answer this: there are too many
factors. Clearly, it is not just the shape of the neuron, but
where on that shape other neurons synapse that determine how
it reacts.
> Finally, has anyone determined what role, if any, DNA might play in
> the processing performed by a neuron?
Not much is known (at least to me), though some experiments have shown
that during imprinting in chicks there seems to be an increase in RNA
synthesis in part of the brain, followed by an increase in protein synthesis.
This seems to implicate DNA in learning, but care must be taken with these
results, as it is hard to distinguish effects due to learning from
effects due to general stimulation of the organism. There is also
work (by Agranoff) that shows inhibition of protein synthesis effects
learning in goldfish in interesting ways.
------------------------------
Subject: SBIR on parallel processing
From: ohare@itd.nrl.navy.mil (John O'Hare)
Date: Wed, 14 Dec 88 09:03:04 -0500
1. Researchers in small businesses (less than 500 people) might be
interested in participating in a research program on acoustic classification
with parallel-processing networks. Awards are $50K for a 6-month definition
phase; and in later competition, up to $250K for each of two years in the
work phase. Close date is 6 Jan 89.
2. The topic is #N89-003 (pg. 87) in the DoD program solicitation entitled
FY-89 Small Business Innovation Research (SBIR) Program. The general contact
is: Mr. Bob Wrenn, SBIR Coordinator, OSD/SADBU, US Dept of Defense,
Pentagon, Rm. 2A340,Washington, DC. 20301-3061. Phone: (202) 697-1481.
------------------------------
Subject: Question for you or your viewers
From: pluto%cs@ucsd.edu (Mark E. P. Plutowski)
Date: Wed, 14 Dec 88 13:16:34 -0800
Re: Radial basis functions and similar items
Last spring I created unit-assemblies, which I then used as nodes. One is,
according to the terminology used in your recent articles, a
spherically-graded unit; the other would be most accurately called an
ellipsoidally-graded unit. My implementation is such that the location and
parameters of the surface are explicitly represented. These assemblies are
of low complexity, and learn via back-prop.
I submitted this news to NIPS, but alas, perhaps due to lack of meat in my
hastily submitted summary and abstract, it was not accepted. I have not
submitted this to anything else since, due to other obligations.
My question is: in light of the recent discussions of spherically graded
units, is what I did now Old Hat?
thanks, keep up the good work!
Mark Plutowski INTERNET: pluto%cs@ucsd.edu
Department of Computer Science, C-014 pluto@beowulf.ucsd.edu
University of California, San Diego
La Jolla, California 92093
------------------------------
Subject: B.P. nets as associative memory nets.
From: nadi@janus.berkeley.edu (Fariborz Nadi)
Organization: University of California, Berkeley
Date: 15 Dec 88 00:47:22 +0000
Fellow neural-neters:
I am presently doing modeling of microfabrication processes using
neural-nets, specifically back-propagation nets to learn the nonlinear
relation between the input and output nodes. This is the first part of the
story, Now having a model I would like to use an associative memory type
network to learn the groups of the input output pairs chosen by an expert.
This is to divide the space of the model into subspaces that are interesting
to an expert in terms of some optimal choices in his/her mind. The second
net will help a novice make close to optimalchoices for a given
partial-input partial-output pair.
I am trying to use again a second back-propagation net as an associative
type net. The way it works , I will use a net that as input has the input
and output of the first net, and as the output has the same, Therefore
creating a mapping between similar patterens. Kind of like 8-3-8 network
(coding decimal to binary back to decimal). After the network has learned
the mapping given a partial input and partial output( the same ) we can lock
the values of these nodes and holding the weights and thresholds constant,
change the values of the unknown input and output nodes, given that the
corresponding input-output nodes should change together. This can be done
using an optimization technique, not necessarily a locally computable one.
Now I have two questions:
1) What do you think about the use of backpropagation nets as an associative
type net, given the method discribed above or some other?
2) Is this work done before and if so where is it published? What would be very
interesting to me is finding an optimization technique for the second part
that would be locally computable.
------------------------------
End of Neurons Digest
*********************
