SuperHack/SuperHack CD.bin/CODING/MISC/NUD-0104.ZIP/NUD-0104.TXT

Internet Message Format | 1992-11-29 | 20.9 KB

Return-Path: @ti.com:[e-mail]
Received: from hplabs.HP.COM by hplpm.HPL.HP.COM; Fri, 12 Aug 88 17:07:12 pdt
Received: from RELAY.CS.NET by hplabs.HP.COM with SMTP ; Fri, 12 Aug 88 16:06:37 PST
Received: from ti.com by RELAY.CS.NET id aa01304; 9 Aug 88 14:52 EDT
Received: by ti.com id AA19108; Tue, 9 Aug 88 13:05:46 CDT
Message-Id: <[e-mail]>
Received: by tilde id AA02920; Tue, 9 Aug 88 12:53:04 CDT
Date: Mon, 22 Dec 86 12:53:04 CDT
From: "Michael T. Gately" <[e-mail]>
To: [e-mail], [e-mail]
Subject: NEURON Digest V1 / #4


NEURON Digest       22 DEC 1986       Volume 1 Number 4

Topics in this digest --
   Queries - Neurocomputer references &
             Text-to-speech conversion
   Replies - ICNN Questions
   News - Project Proposal
   Seminars/Courses - Connectionism and Cog. Linguistics
   Long Messages - Physics, Dynamical systems, and Neural Networks &
                   Challenge to Connectionists &
                   Electromagnetic vs. Electrical signalling

-----------------------------------------------------------------

From:    [e-mail] 22-DEC-1986 08:09
To:    [e-mail]
Subj:    Request for references on Neurocomputers

I have recently seen some mention of neurocomputers and am curious as to
what they really are; what level of the biological system they are trying
to model, as well as the implementation.  If anybody can point me to references
on the subject, or descriptions of present projects, it would very much be
appreciated.  I'll summarize on the net if enough people express an interest.

Thank you

-charles

[e-mail]              Box 1662
iuvax!cdaf                    Bloomington IN 47402-1662
[e-mail]        [phone]

-----------------------------

From:    [e-mail] 22-DEC-1986 08:04
To:    [e-mail]
Subj:    Text-to-speech conversion


    I am a graduate student in UC Santa Barbara. At present I am
writing a dissertation on text-to-speech conversion using a neural network
model (similar to the NETtalk experiment conducted by Sejnowski & Rosenberg.)

    I would like to get information about people working on text-to-speech
 conversion projects using different approaches.

        Thanks.

                - Umesh D. Joglekar

     e-mail : [e-mail]

    USnail  : Umesh D. Joglekar
          Mail Stop 230-5
          NASA Ames Research Center
          Moffett Field, Ca 94035

              [phone]

-----------------------------

From: [e-mail] 22-DEC-1986 10:41
Reply to: [e-mail]
Subj: First Annual International Conference on Neural Networks

In reply to the questions I got concerning the message in the
last digest about the ICNN;

A) I do not know of any reduced fee for students, call Nomi
   Feldman (address below).

B) I do not knwo if the $250 before Jan 31 contains proceedings,
   call Nomi Feldman (address below).

C) Checks are to be made out to IEEE First Annual ICNN, care
   of Nomi Feldman (address below).

D) I do not know if there was an official Call for Papers, or
   if all the contributors have already been selected.  I suggest
   you call Maureen Caudill, ICNN, 10615G Tierransanta Blvd.,
   Suted 346, San Diego, CA  92124.

E) If you have any other questions, contact Nomi Feldman (address
   follows).

Nomi Feldman
Conference Coordinator
3770 Tansy Street
San Diego, CA  92121
[phone]

-----------------------------

From:    [e-mail] 22-DEC-1986 08:09
To:    neuron@ti-csl
Subj:    Project Proposal


This is an idea I have for a project employing Connectionist networks.
Please send me any comments you might have as I am in need of feedback.

I would like to develop a system that would take a stream of notes and
determine the best way to play it on the guitar. This is not a trivial
problem as there are many possible ways of playing a given note. To decide
how to play something, decisions must be based on the current position
of the fingers, which fingers can reach where, what is coming up, and what
there has been in the past (to predict what might be coming in the future).

The main goal of the project is to build an application with a "Connectionist
Model" (ballard and feldman, 1982). The guitar transcriber may not be the
most fascinating project, but I think it is feasable and different from
what has been done. The ability for connectionist nets to deal with multiple
constraints would be exploited by such a project.

I am especially interested if something like this has been done before.
What other types of problems have been solved with connectionist nets?


Michael Lee Gleicher            (-: If it looks like I'm wandering
    Duke University            (-:    around like I'm lost . . .
Now appearing at : duke!ravi        (-:
Or P.O.B. 5899 D.S., Durham, NC 27706    (-:   It's because I am!

-----------------------------

From:    [e-mail] 22-DEC-1986 08:12
To:    [e-mail]
Subj:    Princeton seminar (past) - Connectionism and Cog. Linguistics

TITLE:     Connectionism and Cognitive Linguistics
SPEAKER:   George Lakoff, University of California, Berkeley
DATE:      Monday, December 8
LOCATION:  Princeton University, Green Hall, Langfeld Lounge
TIME:      12 Noon

In the 1970's Cognitive Science developed largely under the assumption that
human reason could be characterized in terms the manipulation of symbols
that were to get their meaning via a relation to things in the world. This
view grew out of the attempt to use mathematical logic and model theory as a
basis for the study of human reasoning. Over the past decade it has become
increasingly clear that such an approach cannot work. In its place there has
developed a cognitive approach to semantics.  This talk will (1) provide an
overview of the phenomena that have led to the development of cognitive
semantics, (2) survey the mechanisms used in cognitive semantics, and (3)
discuss how such mechanisms might be made sense of within the emerging
connectionist theory of mind.

-----------------------------

From:    [e-mail] 22-DEC-1986 08:13
To:    [e-mail]
Subj:    Physics, Dynamical systems, and Neural Networks

Included below is a message I sent to the Arpa physics mailing-list
a while back. Are people on this list interested in similar topics?

Should this mailing-list be called complex-systems?

There is a new journal coming out called Complex Systems that deals
with topics that are relevant to this list. The subscription price
for individuals is $65. Send payment to:

    Complex Systems Publications
    P.O. Box 6149
        Champaign, IL 61821-8149

The University of Illinois is also starting a center for complex
systems. Stephen Wolfram is the director.

There is an article on chaos in the December Scientific
American. The authors of this article formed a group while they
were at UCSC, called the Dynamical Systems Collective, and there
is a book out on their exploits to beat the Las Vegas casinos
using dynamical-systems theory called "The Eudaemonic Pie" by
Thomas Bass.

Begin message:

-----------------------
I have some remarks to make on the recent discussion on
quantum mechanics, determinism, and randomness.

This also provides me the opportunity to suggest several topics
for discussion that have not been discussed here and are of
current interest to researchers in physics, computer science,
and mathematics:


   - Complex (nonlinear) dynamical systems.

   - Quantum Computers.

   - Neural-network-like architectures  for  optimization.  (Borrows  from
     work in ill-condensed matter such as Ising models of spin glasses).

   - Cellular automata models of physical systems.

   - Edward Nelson's theory of quantum fluctuations, or the deBorglie-Bohm
     model of quantum mechanics which postulates nonlocal potentials.


To start things off, I will mention several articles and use
them to reply to a couple of statements that have been made on
this list:

[Deutsch 85]   Deutsch, D.  Quantum Theory, the Church-Turing Principle and the
               Universal Quantum Computer.  Proc. R. Soc. Lond. A400:97-117,
               1985.

[Erber 85]     Erber, T. and S. Putterman.  Randomness in Quantum Mechanics -
               Nature's Ultimate Cryptogram?  Nature 318:41-43, November 7,
               1985.

[Ford 83]      Ford, Joseph.  How Random is a Coin Toss?  Physics Today :40-47,
               April, 1983.

[Hopfield 86]  Hopfield, John J., & David W. Tank.  Computing with Neural
               Circuits: A Model.  Science , 8 August, 1986.

[Kirkpatrick 83]
               Kirkpatrick S., C.D. Gelatt and M.P. Vecchi.  Optimization by
               Simulated Annealing.  Science 220(4598):671-680, May 13, 1983.

[Marroquin 85] Marroquin, Jose Luis.  Probabilistic Solution of Inverse
               Problems.  Technical Report AI-TR 860, MIT AI Lab, September,
               1985.

[Wolfram 86]   Wolfram, Stephen.  Origins of Randomness in Physical Systems.
               In Wolfram, Stephen (editor), Theory and Applications of
               Cellular Automata, pages 298-301. World Scientific Publishing Co,
               Singapore, 1986.

Ford's article is a good introduction to the exciting field of
nonlinear dynamics. A point Ford makes in the paper is that it
becomes hard (actually in a computer science sense) for us to
distinguish between a system that is nondeterministic and one
that is deterministic but nonlinear and chaotic. Nonlinear
chaotic systems take small perturbations and explode them. We
can actually consider a notion of deterministic randomness! Note
that I do *not* mean pseudo-randomness - this arises in computer
random number generators because of the finite word size.  I
have read this article several times and still marvel at what it
says.  [I think it would have been an interesting twist of
events if quantum mechanics was developed *after* our present
understanding of nonlinear systems.]

The second article is a controversial introduction to the
current work in quantum computers. Deutsch's claim is that
quantum computers are more powerful than Turing machines because
they would be nondeterministic.

Erber's article suggests an experiment that would determine
whether quantum systems are truly random or pseudorandom by doing
cryptographic analysis of the fluorescence of an isolated atom.
(The ability to isolate single atoms is an exciting new
experimental technique.)

Kirkpatrick's article discusses simulated annealing which
incorporates random numbers as a key to its workings. I offer
this and the the subject of Tank's article as examples of how random
numbers can be useful computation including computation in
brains. I offer this in response to:

  From: "Keith F. Lynch" <KFL%MX.LCS.MIT.EDU@@MC.LCS.MIT.EDU>
    Uncompensated quantum effects would act as a random number generator.
  Adding data from such a "generator" would not help the brain reach any
  sort of optimal decision, since the output of the "generator" would not
  correlate with any factors relevant to the task at hand.

An understanding of why random numbers can be useful can be had
if one appreciate the fact that many problems are inverse
problems with many possible solutions.  These problems are
*ill-posed* mathematically.  We are left with a search through a
large space of possibilities. The search space has many local
minima and random numbers are used as noise to escape local
minima. An example inverse problem the brain must deal with:
finding the objects and their relationships within the visual
field. The Marroquin's reference is a thesis on this subject.

The article by Wolfram discusses the sources of randomness in
nature. One source is the fact that many complex systems are
open and are immersed in a "heat bath". In fact this can be a
good source of random numbers for distributed or parallel
computers.  (A typical reason why distributed computer networks
are "open" is that the processor clocks of machines on a network
are not synchronized.  We either have to have synchronization
primitives in our programming language - which is the common
means of handling this case - or we have to invent ways to
program with a certain amount of "indeterminism".)  The
understanding of complex behavior of computer networks requires
a new set of tools than what we in computer science have been
used to. This is my present area of interest.  These "open
system" parallel models of computation are probably more
appropriate for the brain than are serial "closed system"
models.


  From: "Keith F. Lynch" <KFL%MX.LCS.MIT.EDU@@MC.LCS.MIT.EDU>
 But nothing OBSERVABLE about their effects is any different
 than a random number generator, if we are talking about non-
 deterministic effects.  By "quantum effect" I include only these.
 I do not include things such as tunnel diodes or their equivalent (if
 any) in the brain.  If I were to include those and other deterministic
 devices which rely on quantum mechanics to work, I would have to
 include the whole universe as a "quantum effect".

A closing observation: We all tend to confuse reality with
models of reality. Quantum effects such as tunneling are
empirically observable. Nondeterminism of quantum mechanics
belongs to the model. In the light of deterministic randomness,
how could one test for nondeterminism? I have been collecting
evidence that asynchronous parallel computation in fact exhibits
quantum effects such as tunneling and observer-observed
interaction. By the way, this does not have to be a hidden
variable type of interpretation. Time slips in the communication
of information across processors deals with the topology of
simulated space-time that the computation is embedded in.

Albert Boulanger
BBN Labs

-----------------------------

From:    HARNAD%MIND@PRINCETON 22-DEC-1986 08:12
To:    neuron@ti-csl
Subj:   Challenge to Connectionists

I would like to issue a challenge to connectionists. Connectionist (C)
approaches are receiving a great deal of attention lately, and many
ambitious claims and expectations have been voiced. It is not clear,
on the existing evidence, what the null hypothesis is or ought to be,
and what would be needed to reject it. Let me propose one:

H-0: Connectionist approaches will fail to have the power to capture
the capacities of the mind because they will turn out to be subject to
higher-order versions of the same limitations that eliminated
Perceptrons from contention.

It would seem that in order to reject H-0, meeting one or the other of the
following criteria will be necessary:

    (i) Prove formally that not only is C not subject to perceptron-like
    constraints, but that it does have the power to generate
    mental capacity.

This first criterion is currently rather vague, since there is no well-defined
formal problem that is known to be equivalent to mental capacity (in the way
the traveling salesman problem is known to be equivalent to many important
computational problems). The conceptual and evidential burden,
however, is on those who are making positive claims.

    (ii) Demonstrate C's power to generate mental capacity empirically
    by generating human performance capacity or a significant portion
    of it.

The second criterion also suffers from some vagueness because there
seems to be no formal, empirical or practical basis for determining
when (if ever) a performance domain ceases to be a "toy" problem (like
chess playing, circumscribed question-answering and
object-manipulation, etc.) and becomes life-size -- apart from the
Total Turing Test, which some regard as too demanding. It is also
unknown whether there exists any natural (or formally partitionable)
subtotal performance "module." Again, however, the conceptual and
evidential burden would seem to be on those who are making positive
claims.

To summarize, my challenge to connectionists is that they either
provide (1) formal proof or (ii) empirical evidence for their claims
about the present or future capacity of C to model human performance
or its underlying function.

Conspicuously absent from the above is any mention of the brain. The
brain is a red herring at this stage of investigation. Experimental
neuroscientists have only the vaguest ideas about how the brain
functions. They, like all other experimental scientists, must look to
theory not only for hypotheses about function, but for guidance as to
what to look for. There is no reason to believe, for example, that the
functional level "where the action is" in the brain is anything
remotely similar to our naive and simplistic picture consisting of neurons,
action potentials, and their connections. It may, for example, be at the
subthreshold level of graded postsynaptic potentials, or at a biochemical
level, or at no level so far ascertained or even conceptualized.

At this point, taking it to be to C's credit that it is "brain-like"
amounts to the blind leading the blind. Indeed, I would recommend a
"modularization" between the efforts of those who test C as a neural
modal and those who test it as a performance model. The former should
restrict themselves to accounting for the data from experimental neuroscience
and the latter should restrict themselves to accounting for performance data,
with neither claiming the other's successes as bearing on the validity
of their own efforts.  Otherwise, shortcomings in C's performance
capacity will be overlooked or rationalized on the grounds of brain
verisimilitude and shortcomings in C's brain-modeling will be overlooked
or rationalized on the grounds of its cognitive capacity.

Finally, lest it be thought that AI (symbolic modeling) gets off
scot-free in these considerations: AI is and should be subject to the
same two criteria. "Turing power" is no better a prima facie basis for
claiming to be capturing mental power in AI than "brain-likeness" is in
connectionism. Indeed, C has the slight advantage that it is at least
a class of algorithms rather than just a computational architecture.
Hence it has some hope of showing that, what (if anything) it can
ultimately do, it does by the same general means, rather than ad hoc
ones gerrymandered to any problem at hand, as AI does.

Instead of indulging in mentalistic (and in C's case, also neuralistic)
overinterpretations of the minuscule performance capacities of current
models, both AI and C should hunker down to creating performance
models that will require no embellishment or interpretation to be
impressive as inroads on human performance and its functional basis.
--

Stevan Harnad                                  (609) - 921 7771
{allegra, bellcore, seismo, rutgers, packard}  !princeton!mind!harnad
[e-mail]

-----------------------------

From:    [e-mail] 22-DEC-1986 08:12
To:    [e-mail]
Subj:    Electromagnetic vs. Electrical signalling

Let me attempt to stir some controversy:

E. Roy John, at NYU Medical Center, is one proponent of the idea that
information processing in the brain is mediated by electromagnetic,
rather than electrical, signals.  In the magazine "High Technology",
August 1984, ("Why Can't a Computer be More Like a Brain?"), an experiment
with a cat is offered as supporting evidence:

    A cat is taught to turn left at a T intersection when presented
with a 2Hz tone, and to turn right when presented with a 4Hz tone.
Later, the experiment was repeated using either a 2Hz or 4Hz signal applied
to stimulation electrodes inserted in the cat's brain.  The cat, as would
be expected, turned in the appropriate directions.  John concludes that,
since the electrodes are too big to stimulate any discrete pathway,
that the neurons in the brain act cooperatively.  This seems reasonable,
so far.

    He then fed 2Hz signals to each of two stimulation electrodes,
with the two signals being out of phase.  The cat behaved as would be
expected for a 4Hz signal.  John then concludes that the electromagnetic
field, as a whole, is responsible for this phenomenon.  Here is where I differ?

(Flame on!) There is no reason to suppose that the electromagnetic field is
any more responsible than the simple melding of the electrical signals.
It seems to me that John is being mislead, by the electromagnetic
nature of EEG data, into believing some fantastic explanation.  Never
depend on a complicated explanation when a simple explanation will suffice:
ie. the commonly accepted theory of electrical pulses on neural axons being
the primary means of communication (Flame off.)

Does anyone have any more information, or a better understanding of this
question?  I'd like to hear about it.

Brian  (Lantz@Rutgers)

-----------------------------

End of NEURON Digest
********************