Copy Link
Add to Bookmark
Report
Neuron Digest Volume 09 Number 23
Neuron Digest Monday, 1 Jun 1992 Volume 9 : Issue 23
Today's Topics:
Administrivia
CEDAR CDROM 1: Handwritten words and characters
Fuzzy-logic newsletter via e-mail?
1993 Connectionist Summer School preliminary announcement
Announcing the availability of a hyperplane animator
Question in IA-Nets
looking for SunNet Ref. Manual
Attack on Scientific grants
JOB AD: Hybrid Hidden Markov/Connectionist Models for Speech Recognition
Send submissions, questions, address maintenance, and requests for old
issues to "neuron-request@cattell.psych.upenn.edu". The ftp archives are
available from cattell.psych.upenn.edu (128.91.2.173). Back issues
requested by mail will eventually be sent, but may take a while.
----------------------------------------------------------------------
Subject: Administrivia
From: "N-D Moderator, Peter Marvit" <neuron@cattell.psych.upenn.edu>
Date: Mon, 01 Jun 92 12:04:14 -0500
A brief apology for the unexplained absence of the Neuron Digest for the
past few weeks. A small (!) project at the end of the semester plus a
move to a new house took much more time than I had expected. Thus,
Neuron Digest (plus a few other activities) went on an unannounced
holiday.
Regular issues will resume now. Thank you, as always, for your
readership and patience.
-Peter Marvit
Neuron Digest Moderator
------------------------------
Subject: CEDAR CDROM 1: Handwritten words and characters
From: hull@sybil.cs.Buffalo.EDU (Jon Hull)
Organization: State University of New York at Buffalo/Comp Sci
Date: 23 Apr 92 16:42:39 +0000
The Center Of Excellence for
Document Analysis and Recognition (CEDAR)
State University of New York at Buffalo
announces the availability of
CEDAR CDROM 1:
USPS Office of Advanced Technology Database of Handwritten
Cities, States, ZIP Codes, Digits, and Alphabetic Characters
CEDAR is pleased to announce the availability of a
CDROM database that contains handwritten words and ZIP Codes
in high resolution grayscale (300 ppi 8-bit) as well as
binary handwritten digits and alphabetic characters (300 ppi
1-bit). This database is intended to encourage research in
off-line handwriting recognition by providing access to
handwriting samples digitized from envelopes in a working
post office.
Specifications of the database include:
+ 300 ppi 8-bit grayscale handwritten words (cities,
states, ZIP Codes)
o 5632 city words
o 4938 state words
o 9454 ZIP Codes
+ 300 ppi binary handwritten characters and digits:
o 27,837 mixed alphas and numerics segmented
from address blocks
o 21,179 digits segmented from ZIP Codes
+ every image supplied with a manually determined
truth value
+ all data digitized on an Eikonix EC850 CCD scanner
+ extracted from live mail in a working U.S. Post
Office
+ simulates a "real" recognition environment:
o unrestricted for author
o unrestricted for writing style
o unrestricted for writing implement (pen, pen-
cil, ...)
o authors had no knowledge their samples were
to be used
+ divided into explicit training and test sets (90%
training and 10% testing)
+ performance comparison between researchers possi-
ble on the same data sets.
+ word images in the test set supplied with dic-
tionaries of postal words that simulate partial
recognition of the corresponding ZIP Code.
+ digit images included in test set that simulate
automatic ZIP Code segmentation. Results on these
data can be projected to overall ZIP Code recogni-
tion performance.
+ image format documentation and software included
Suitable for automated handwritten word recognition
research, the database can be used for:
o algorithm development
o system training and testing
The database is a valuable tool for developing the
range of techniques needed for high performance handwritten
word recognition including preprocessing, segmentation,
feature extraction, and classification. The system require-
ments are a 5.25" CD-ROM drive with software to read ISO-
9660 format.
For any further information, including how to order the
database, please contact:
Jonathan J. Hull
Associate Director, CEDAR
226 Bell Hall
State University of New York at Buffalo
Buffalo, NY 14260
716-636-3195 (voice)
716-636-3966 (fax)
hull@cs.buffalo.edu (email)
Please note: email contact is preferred and is most efficient
------------------------------
Subject: Fuzzy-logic newsletter via e-mail?
From: James Rash <jim@class.gsfc.nasa.gov>
Date: Sun, 03 May 92 12:03:25 -0500
[[ Editor's Note: I'm sure there must be something, but I don't know of
any fizzy e-lists. Can one of your readers help, and send a copy of your
reply to Neuron Digest? -PM ]]
I am looking for a fuzzy-logic newsletter available via e-mail. Any info
to help find one would be appreciated.
Jim Rash/Code 531.1
NASA Goddard Space Flight Center
Greenbelt, MD 20771
301-286-3595
------------------------------
Subject: 1993 Connectionist Summer School preliminary announcement
From: "Michael C. Mozer" <mozer@dendrite.cs.colorado.edu>
Date: Sun, 03 May 92 11:50:02 -0700
PRELIMINARY ANNOUNCEMENT
CONNECTIONIST MODELS SUMMER SCHOOL
JUNE 1993
University of Colorado
Boulder, CO
The next Connectionist Models Summer School will be held at the
University of Colorado, Boulder in the summer of 1993, tentatively June
23-July 6. This will be the fourth session in the series that was held
at Carnegie-Mellon in 1986 and 1988, and at UCSD in 1990.
The Summer School will offer guest lectures and workshops in a variety of
areas of connectionism, with emphasis on theoretical foundations,
computational neuroscience, cognitive science, and hardware
implementation. Proceedings of the Summer School will be published the
following fall.
As in the past, participation will be limited to graduate students
enrolled in PhD programs (full or part time). We hope to have sufficient
funding to subsidize tuition and housing.
This is a preliminary announcement. Information about how to apply will
be posted over connectionists. As a point of contact, address electronic
correspondence to "cmss@cs.colorado.edu".
Jeff Elman
University of California, San Diego
Mike Mozer
University of Colorado, Boulder
Paul Smolensky
University of Colorado, Boulder
Dave Touretzky
Carnegie-Mellon University
Andreas Weigend
Xerox PARC and University of Colorado, Boulder
------------------------------
Subject: Announcing the availability of a hyperplane animator
From: pratt@cs.rutgers.edu
Date: Mon, 04 May 92 18:01:05 -0500
-----------------------------------
Announcing
the availability of an
X-based neural network hyperplane animator
-----------------------------------
Lori Pratt and Paul Hoeper
Computer Science Dept
Rutgers University
Understanding neural network behavior is an important goal of many
research efforts. Although several projects have sought to translate
neural network weights into symbolic representations, an alternative
approach is to understand trained networks graphically. Many
researchers have used a display of hyperplanes defined by the weights
in a single layer of a back-propagation neural network. In contrast to
some network visualization schemes, this approach shows both the
training data and the network parameters that attempt to fit those
data. At NIPS 1990, Paul Munro presented a video which demonstrated
the dynamics of hyperplanes as a network changes during learning. This
video was based on a program implemented for SGI workstations.
At NIPS 1991, we presented an X-based hyperplane animator, similar
in appearance to Paul Munro's, but with extensions to allow for
interaction during training. The user may speed up, slow down, or
freeze animation, and set various other parameters. Also, since it
runs under X, this program should be more generally usable.
This program is now being made available to the public domain. The
remainder of this message contains more details of the hyperplane
animator and ftp information.
- ------------------------------------------------------------------------------
1. What is the Hyperplane Animator?
The Hyperplane Animator is a program that allows easy graphical display
of Back-Propagation training data and weights in a Back-Propagation
neural network.
Back-Propagation neural networks consist of processing nodes
interconnected by adjustable, or ``weighted'' connections. Neural
network learning consists of adjusting weights in response to a set of
training data. The weights w1,w2,...wn on the connections into any one
node can be viewed as the coefficients in the equation of an
(n-1)-dimensional plane. Each non-input node in the neural net is thus
associated with its own plane. These hyperplanes are graphically
portrayed by the hyperplane animator. On the same graph it also shows
the training data.
2. Why use it?
As learning progresses and the weights in a neural net alter, hyperplane
positions move. At the end of the training they are in positions that
roughly divide training data into partitions, each of which contains only
one class of data. Observations of hyperplane movement can yield
valuable insights into neural network learning.
3. How to install the Animator.
Although we've successfully compiled and run the hyperplane animator on
several platforms, it is still not a stable program. It also only
implements some of the functionality that we eventually hope to include.
In particular, it only animates hyperplanes representing input-to-hidden
weights. It does, however, allow the user to change some aspects of
hyperplane display (color, line width, aspects of point labels, speed of
movement, etc.), and allows the user to freeze hyperplane movement for
examination at any point during training.
How to install the hyperplane animator:
1. copy the file animator.tar.Z to your machine via ftp as follows:
ftp cs.rutgers.edu (128.6.25.2)
Name: anonymous
Password: (your ID)
ftp> cd pub/hyperplane.animator
ftp> binary
ftp> get animator.tar.Z
ftp> quit
2. Uncompress animator.tar.Z
3. Extract files from animator.tar with:
tar -xvf animator.tar
4. Read the README file there. It includes instructions for running
a number of demonstration networks that are included with this
distribution.
DISCLAIMER:
This software is distributed as shareware, and comes with no warantees
whatsoever for the software itself or systems that include it. The
authors deny responsibility for errors, misstatements, or omissions that
may or may not lead to injuries or loss of property. This code may not
be sold for profit, but may be distributed and copied free of charge as
long as the credits window, copyright statement in the ha.c program, and
this notice remain intact.
------------------------------
Subject: Question in IA-Nets
From: Ulf Rimkus <RIMKUS_U%DMRHRZ11.BITNET@vm.gmd.de>
Date: Wed, 06 May 92 22:41:10 +0700
Dear Neural-Netters!
Just now I am going to develop a neural net similar to the IA-Net by
Rumelhart and McClelland. Now I am searching for the "weight-matices" you
can find in the formulas.
------ ------
\ \
\ \
n (t)= / alpha e (t) - / gamma i (t)
i / ij j / ik k
------ ------
j k
But if you look into the sources given by Rumelhart and McClelland in
their book "Explorations in parallel distributed processing" (Chapter 7)
those weight-matrices 'alpha' and 'gamma' are difficult to find. I only
could find a _vector_ 'alpha' and 'gamma'.
My aim will be to go to a stochastic net and/or to replace the IA-Net by
a Boltzmann-Machine as it is suggested by McClelland(1991). And that's
why I think it could be nice to have these weight-matrices explicitly.
My hope now is, that there is someone in the net, who already had worked
out how the weights are built up and would like to share his knowledge
with me.
Yours,
Ulf Rimkus.
INTERNERT: rimkus_u@dmrhrz11.hrz.uni-marburg.de-----------------------------
Hey Fremder, meine besten Freunde sind Fremde, aber Du warst noch nie hier!
BITNET: rimkus_u@dmrhrz11---------------------------------------------------
------------------------------
Subject: looking for SunNet Ref. Manual
From: Pablo Dominguez <fibces04@esaii.upc.es>
Date: Wed, 06 May 92 19:07:50 +0100
We are working with the Sunnet neurosimulator. We'd like to get the
Reference Manual. If someone has it, please, let us know.
Thanks in advance
Pablo Dominguez
Dept ESAII (UPC)
Barcelona, Spain
E-mail: fibces04@esaii.upc.es
------------------------------
Subject: Attack on Scientific grants
From: hyman@garnet.berkeley.edu
Date: Fri, 08 May 92 16:21:23 -0800
[[ Editor's Note: This message is United States specific. Although the
issues involved are broader than neural nets, this message may be worth
your careful consideration since it may apply to other fields in the
future. My cordial thanks to the current President and members of
Congress for making this note necessary. -PM ]]
I am passing on the following distressing message, which I just
received:
Date: Tue, 5 May 1992 15:16:27 EDT
[Reply-To]: APA Research Psychology Network <APASD-L@VTVM2.BITNET>
[Sender]: APA Research Psychology Network <APASD-L@VTVM2.BITNET>
[from] Cheri Fullerton <APASDCF@GWUVM.BITNET>
Subject: ACTION ALERT - Psychology Funding Cuts!
PUBLIC POLICY OFFICE
ACTION ALERT
[to] APA Funding Bulletin Recipients
[from] Barbara J. Calkins
Associate Director, Public Policy Office
DATE: May 5, 1992
RE: Threatened Cuts to Funded Research Programs &
Attack on Peer/Merit Review
Last week the Senate Appropriations Committee passed a bill taking
back, or "rescinding" in Washington jargon, $8.3 billion in
already-approved Fiscal Year 1992 funding. The rescissions included
specific grants, identified by title, that had already been peer-reviewed
and, in some cases, awarded by federal agencies including the National
Science Foundation (NSF) and the National Institutes of Health (NIH).
Related legislation developed by the House of Representatives
Appropriations Committee sets broad rescission targets, allowing federal
agencies the latitude to take cuts from new and as yet unfunded programs
rather than reneging on commitments to ongoing research. The Senate bill
poses serious problems, and WE NEED YOUR HELP NOW.
BACKGROUND
The most frightening aspect of the Senate bill is its direct
assault to American science's peer/merit review system. Heretofore it
has been the business of scientists, not Congress, to decide the merits
of individual grant applications. Such decisions have traditionally had
a scientific basis, and have been made by other researchers who carefully
consider each proposal. The Senate, on the other hand, has apparently
combed through lists of grant titles and culled out the ones with
trivial-sounding titles, targeting them for termination. This is an
alarming precedent.
Another alarming precedent set by the Senate's action is the
rescission of funds that have already been committed or awarded.
Scientists use these commitments to make agreements with their supporting
academic institutions, with colleagues, and with graduate students, in
order to ensure a stable and productive research enterprise. While these
particular Senate-proposed rescissions may affect a relatively small
number of scientific investigations, the fact that Congress can
arbitrarily reverse scientifically- based research funding decisions
threatens the stability of the scientific research process in a manner
that could cripple many research programs.
An additional concern is the apparent focus of the Senate
rescissions on behavioral and social science. Many of the grants
targeted by the Senate for rescission are from the new Social, Behavioral
and Economic Sciences Directorate of the National Science Foundation.
Other proposed cuts come at the expense of behavioral and social research
being supported by the National Institutes of Health-- specifically, the
dental pain and fear work of the National Institute of Dental Research.
Clearly, we must protect psychological research by better educating
Members of Congress about its nature and significance. We need your help
in doing so.
HOW YOU CAN HELP
As a constituent, yours is an important voice in the development of
federal science policy and funding decisions. Your contacts to Congress
MAKE A DIFFERENCE. We need you to: 1) Contact your Senators and Member
of Congress and ask them to: Oppose the Senate budget rescissions bill
(S.2403), and the politicization of science. Tell them that if these
rescissions must occur, then the House approach is more appropriate; and
2) Contact Senate and House Appropriations Committee members with the
same message. These individuals are in key positions to influence the
final outcome on this critical issue:
Senate House of
Representatives
Robert Byrd (D-WV) William Natcher(D-KY)
Tom Harkin (D-IA) Carl Pursell (R-MI) Barbara
Mikulski (D-MD) Bob Traxler (D-MI) Brock Adams
(D-WA) Bill Green (D-NY) Slade Gorton (R-WA)
In such contacts you should identify yourself as a scientist and,
if appropriate, as a constituent. You can raise the issues outlined
above to make your point, and you can develop them further with details
of your own experiences. You can make your contact by telephoning, or by
Western Union "Public Opinion Message".
Telephone - You can reach any Congressional office in Washington by
dialing 202/224-3121 and asking to be connected with your legislator's
office. Ask to speak with the staff member who handles research
appropriations. Give your name, affiliation, and the purpose of your
call. At the conclusion of the conversation, offer yourself as a contact
in the future and give your telephone number.
Western Union - Call Western Union at 1-800-325-6000 and ask to
send a "Public Opinion Message." The cost is $9.95 for 20 words or less
and $3.50 for each additional 20 words. Address your message: The
Honorable (Member's Name), U.S. (Senate or House of Representatives),
Washington, D.C. (20510/House or 20515/Senate).
If you need additional information, or help in identifying your
Senators or Members of Congress, then send a BITNET message to:
APASD@GWUVM. Thanks for your help and support.
------------------------------
Subject: JOB AD: Hybrid Hidden Markov/Connectionist Models
for Speech Recognition
From: Tony Robinson <ajr@eng.cam.ac.uk>
Date: Tue, 12 May 92 11:17:24 +0000
[ My apologies if this appears on your screen more than once ]
Cambridge University Engineering Department
Research Assistant in Speech Recognition
The Speech Group expects to appoint an RA for an ESPRIT project on large
vocabulary speech recognition using hybrid HMM/ANN structures.
Candidates will have PhD experience in one of these techniques and an
interest in C programming in a parallel environment, joining a
multi-nation research collaboration. It is expected that the project
will start during the summer 1992, and will last for 15 months with a
possible extension to 36 months. The RA starting date is negotiable.
Salary is age related on RA1A rates. Further information and application
forms can be obtained from Mavis Barber, Cambridge University Engineering
Department, Trumpington Street, Cambridge CB2 1PZ. Closing date for
applications is 1 June 1992.
Further Particulars
The Speech Group expects to appoint an RA for an ESPRIT project on large
vocabulary speech recognition using hybrid HMM/ANN structures. The
consortium for this Basic Research Project is: Lernout & Hauspie Speech
Products, Brussels; INESC, Lisbon; CUED, with ICSI (International
Computer Science Institute) Berkeley as sub-contractor.
The baseline structures are the HMM/MLP pioneered by Bourlard (LHS) and
Morgan (ICSI) and the HMM/recurrent NN (RNN) studied by Robinson (CUED).
These will be nextended in several ways; by incorporating improvements
analogous to those used in state of the art HMM recognisers; by further
development of the theoretical bases of hybrid classification; by the
development of better acoustic features with enhanced speaker and
communication channel robustness; by the development of better training
procedures; by the investigation of fast speaker adaptation in hybrids.
The project also aims to demonstrate real-time recognisers and their
evaluation against international databases such as used by the DARPA
community. A major feature of the project will be the use of the RAP,
ring array processor, developed by ICSI, at each site; with further
development of hardware and software tools by ICSI for the project. The
RAP has 16 digital signal processors with a peak performance of 0.5
GFlps.
Applicants should have PhD or equivalent experience in HMMs or ANNs in
speech recognition. The RAP uses C & C++ in a UNIX environment and
applicants should have relevant experience and an interest in extending
their skills in a parallel environment. The RA at Cambridge will be
mainly responsible for the development of HMM/RNN structures.
It is expected that the project will start during the summer 1992,
initially for a period of 15 months but with an extension to 36 months.
The starting date for the RA is negotiable.
Further details and application forms from Mavis Barber at Cambridge
University Engineering Department, Trumpington Street, Cambridge CB2 1PZ,
mavis@eng.cam.ac.uk
F. Fallside
A.J. Robinson
May 1992
------------------------------
End of Neuron Digest [Volume 9 Issue 23]
****************************************
