Copy Link
Add to Bookmark
Report
Neuron Digest Volume 02 Number 07
NEURON Digest 3 MARCH 1987 Volume 2 Number 7
Topics in this digest --
Queries - Connectionist Simulators
News - AFOSR Announcement
Abstract - Factual and Counterfactual Reasoning by ...
Seminars/Courses - The Connectionist Air Guitar &
A Computational Perspective on Neural Computing (TI) &
Artificial Neural Networks (HP)
----------------------------------------------------------------------
Date: 1-MAR-1987 01:53
From: duke!ravi
Subject: Connectionist Simulators
Can anyone provide me pointers to simulator programs for Connectionist
Networks/ General Value Passing Networks?
I am proposing to build a simulator for a project, but would much prefer
to use an existing one if it were available. The program I am proposing
would take a description of a network and translate it into a program to
simulate it (much like what YACC does with parsers). By having it build
programs, a high degree of generality can be achived as the definitions
of what the nodes and links are can be written in. Also, like YACC, the
simulation compiler would only build the simulation engine, allowing
the user to add any interface he desires.
Any comments would be appreciated. I would be happy to send my notes on
the simulator ideas to anyone interested.
Thanks
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!
------------------------------
Date: 25-FEB-1987 14:44
From: GILES@AFSC-HQ.ARPA
Subj: AFOSR Announcement
PROGRAM ANNOUNCEMENT: NEURAL COMPUTING AND PROCESSING
The Air Force Office of Scientific Research (AFOSR) announces a
new program of support for basic research on the computational
aspects of neural networks.
Research that could yield computational neural models of
information processing, learning, and cognition in complex
biological systems is specifically encouraged. AFOSR is
interested in multidisciplinary theoretical and empirical
approaches. Research focused on neural architectures subserving
learning and cognition or on computational aspects of
neuromorphic structures and systems is also of interest.
Research proposals are now being accepted by AFOSR. All
proposals received before July 1, 1987 will be considered for the
first cycle of support to begin in October. Support from AFOSR
is typically provided as multi-year grants or contracts.
FOR ADDITIONAL INFORMATION CONTACT:
Dr. C. Lee Giles Architectures and Computation
202-767-4931 GILES@AFSC-HQ.ARPA
Dr. William O. Berry Life Sciences
202-767-5021
Dr. Vincent Sigillito Artificial Intelligence
202-767-5028
Dr. John F. Tangney Life Sciences
202-767-5021 TANGNEY@AFSC-HQ.ARPA
AIR FORCE OFFICE OF SCIENTIFIC RESEARCH
BOLLING AFB, BLDG 410
WASHINGTON, DC 20332-6448
------------------------------
Date: 27-FEB-1987 16:22
From: MAD%G.CS.CMU.EDU@C.CS.CMU.EDU
Subj: Abstract: Factual and Counterfactual Reasoning by ...
Submitted to AAAI-87
Factual and Counterfactual Reasoning
by Constructing Plausible Models
Mark Derthick
Department of Computer Science
Carnegie-Mellon University
Pittsburgh, PA 15213
mad@g.cs.cmu.edu
ABSTRACT
Most of the effort AI has put into common sense reasoning has involved
inference by sequential rule application. This approach is most efficient
in well characterized domains, where any valid chain of inference from a set
of observations leads to an acceptable interpretation. In more realistic
cases where there are multiple consistent interpretations that are not
equally good, or where there are no consistent interpretations, it seems
more natural to choose the best alternative based on the interpretations
themselves rather than the chains of inference used to derive them.
Micro-KLONE is a connectionist network which represents a complete model of
a situation, and strives to maintain consistency by parallel application of
constraints. By generating entire scenarios at once, the need to
heuristically rank applicable rules of inference is eliminated. There is
also no automatic cutoff of chains of inference which lead to scenarios
inconsistent with previously held beliefs, so counterfactual reasoning can
be accomplished by the same mechanism as factual reasoning. An example
requiring conclusions to be drawn from inconsistent beliefs is presented.
Micro-KLONE gives the most plausible answer while default logic cannot
discriminate between this and other answers.
------------------------------
Date: 27-FEB-1987 11:18
From: COTTRELL%ICS%SDCSVAX.UCSD.EDU@C.CS.CMU.EDU
Subject: Seminar: The Connectionist Air Guitar
[forwarded from Connectionists digest - MTG]
SEMINAR
The Connectionist Air Guitar:
A Dream Come True
Garrison W. Cottrell
Department of Air Science
Condominium Community College of Southern California
A major problem faced by many Cognitive Scientists has been
the latent desire to be a rock'n'roll star, without the requisite
talent[1]. Recent advances in connectionist learning mechanisms
(Sutton, 1987) have obviated this need. In this work we present
the design for the connectionist air guitar[2] - the first air
guitar to actually produce the notes played.
This work was motivated by the observation that it is not
hard for people to play the songs of their favorite groups on
their internal phonograph[3] (Kosslyn, 1977). Thus the problem
may simply be one of poor mapping hardware. This suggests that
augmentation by cognitive models may be useful. PDP models are
the obvious candidate for this task, given that they are
"neurally-inspired", or "brain-like"[4]. In this talk we present
the first true augmentation of the mind by a connectionist model,
called Neuro-Acoustic Programming.
We use a three-layer system as follows: Electrodes are
placed on the subject's scalp using the International 10-20
system and amplified by Grass 7P511 preamplifiers[5]. These are
the inputs to the hidden units. The output layer is simply a
localist representation of the notes. These are then interfaced
with a standard guitar synthesizer.
In training, the subject listens to Springsteen while "air
guitaring" the lead. The EEG drives the network, resulting in a
set of outputs. This result is then compared to the correct
output (the music teacher signal) at small delta t's using
Sutton's temporal difference method, and the errors are back-
propagated in the usual way. After two albums, the network
learns to produce the desired notes from the EEG. Of note here
is that the hidden units develop a distributed encoding of the
qualia of the notes, including coarsely-coded features sufficient
to distinguish Jerry Garcia from Conway Twitty[6]. However,
myogram noise in the EEG often leads to noise in the output, so
it appears necessary to implant arrays of silicon electrodes
(developed by Jim Bower at CalTech) directly into the temporal
lobes, eliminating interference from muscle signals. In this
case, the network must actually be borne to run.
____________________
[1]One approach is to ignore this and form a band anyway.
People who took this tack started the punk movement.
[2]An air guitar is a conceptual representation of a guitar,
played in synchrony with actual music. A cult has formed around
this endeavor, with many contests currently being held in local
bars.
[3]Some people claim that they actually can't play the songs
internally as well as they hear them. This is the "bad cognitive
needle" problem, or, in the case of Kosslyn's more advanced
internal cassette player model, "air heads." As long as the sig-
nal uniquely specifies the song, it still maps to the right
notes, so this technique is useful for the hard of thinking.
[4]This is to be contrasted with "neurally-expired", or
"brain-dead" models.
[5]Other types of Grass amplifiers produce a more "sixties-
like" sound.
[6]Some hidden units convert six into nine, the so-called Jimi
Hendrix units (Easy Rider, 1969).
------------------------------
Date: 2-MAR-1987 09:18
From: MOORE%RESBLD@TI-CSL.CSNET
Subject: Colloquium: A Computational Perspective on Neural Computing (TI)
******* RESEARCH COLLOQUIUM *******
SPEAKER: DR. ANDREW G. BARTO - UNIV OF MASS.
TOPIC: A COMPUTATIONAL PERSPECTIVE ON NEURAL
COMPUTING
DATE: FRIDAY, MARCH 6, 1987 3:00 PM
PLACE: Texas Instruments - Expressway Site - Dallas
CENTRAL RESEARCH LABORATORY CAFETERIA
Neural computing (or connectionism) is currently generating widespread
attention, much of which is richly deserved because this approach offers
strengths that are complementary to those of symbolic Artificial
Intelligence. Similarly, connectionist approaches to cognitive modeling
are justifiably attracting attention because of the new types of
cognitive theories they provide. However, the mystique attached to
neural computing due to its association with the wonders of the brain
has a tendency to obscure more mundane links to conventional engineering
methods. In this talk, I focus on connectionist methods that are
intended to provide new solutions to engineering problems. I discuss a
range of connectionist ideas (including some of those on which I have
worked) in the most conventional mathematical and engineering terms
possible. This exercise will show that some connectionist methods are
distinguished mainly by virtue of their implicit (or sometimes explicit)
call for parallel hardware, whereas others go beyond orthodox methods
in deeper ways. I emphasize the importance of trying to work out the
links to more conventional engineering methods so that the genuine
strengths of connectionist ideas can be developed effectively.
BIOGRAPHICAL DATA
Andrew G. Barto received the BS degree in Mathematics and the PhD
degree in Computer Science from the University of Michigan, Ann Arbor,
in 1970 and 1975, respectively. His doctoral dissertation concerned
cellular automata as models of natural systems and initiated an
interest in distributed representation of information. He joined the
faculty of the Dept. of Computer and Information Science at the Univ.
of Mass., Amherst, in 1982 and is currently an Associate Professor.
His research interests are in learning in natural and artificial
systems; connectionist architectures; cooperativity in distributed
systems; models of sensorimotor learning; learning control systems;
models of animal learning and its neural basis.
Members of the Metroplex Institute for Neural Dynamics (MIND) or
other non-TIers interested in attending should present themselves
to the lobby of the Research Building at 2:30 to be given visitor
badges.
------------------------------
Date: 3-MAR-1987 12:17
From: not(LAWS@SRI-STRIPE.ARPA)
Subject: Colloquium: Artificial Neural Networks (HP)
HEWLETT-PACKARD LABORATORIES
COMPUTER COLLOQUIUM
Speaker: David B Parker
Software Consultant
Subject: ARTIFICIAL NEURAL NETWORKS
Recent technological and mathematical advances make it possible to
create electronic circuits and computer programs that behave in many
ways like the human brain. Multilayered networks of artificial
neurons can learn such things as how to produce and recognize speech,
how to correct telescope mirrors for atmospheric disturbances, and
how to control robot arms.
This talk will describe the adaptive algorthms used by artificial
neurons and the algorithms used to connect the artificial neurons
into networks.
Date: Thursday, March 12, 1987
Time: 4:00 pm
Place: Hewlett-Packard, 5M Auditorium
1501 Page Mill Road, Palo Alto
Non HP personnel welcome. Please come to the lobby shortly before 4 pm
so that you may be escorted to the 5M Auditorium.
Refreshments will follow the seminar.
------------------------------
End of NEURON Digest
********************
