Copy Link
Add to Bookmark
Report
Neuron Digest Volume 13 Number 27
Neuron Digest Friday, 22 Apr 1994 Volume 13 : Issue 27
Today's Topics:
Modular NN's and Incremental learning
PhD Studentship Available
Auditory Modelling Software from Loughborough (LUTEar 1.5.0)
Positions for PhD students
postdoc position in NN available
new book
Contents of NETWORK, Volume 5, Number 1.
book listing
Send submissions, questions, address maintenance, and requests for old
issues to "neuron-request@psych.upenn.edu". The ftp archives are
available from psych.upenn.edu (130.91.68.31) in pub/Neuron-Digest or by
sending a message to "archive-server@psych.upenn.edu".
----------------------------------------------------------------------
Subject: Modular NN's and Incremental learning
From: dramesh@research.trddc.ernet.in (D. Ramesh)
Date: Tue, 19 Apr 1994 15:17:59 +0500
Hi everybody,
I am working on Optical character recognition using backpropagation
Neural networks. I am faced with the following problem and would appreciate
any help on the matter:
The Character recognition Neural network that I am developing has to
learn from a huge and changing data base of characters. The character set that
one has to recognise may change depending on the needs of the installation.
The way people write may also be different from installation to installation (
for e.g Indians and European s have a markedly different style of writing).
The problem I am faced with is the prohibitively long training time
for the neural network, because even for a small change in the training and/or
testing data(depending on the needs of the installation which are subject to
change) the neural net has to cycle through all the patterns of the training
set for each learning iteration.Is there any way such as through a modular
approach or incremental training that the neural network tries to acquire only
the knowledge it already does not have, and thus save on training time.
I would welcome any information on modular incremental or other
approaches that may help tackle the above problem.
Regards,
D.Ramesh,
Tata consultancy services,
Pune,
India.
------------------------------
Subject: PhD Studentship Available
From: l.n.kalia@ic.ac.uk
Date: Fri, 15 Apr 1994 13:06:07 -0000
STUDENTSHIP AVAILABLE FOR PhD IN PSYCHOLOGY/NEURAL NETWORKS
===========================================================
Dear Colleagues,
I am passing on the following information on behalf of Dr. Eamon Fulcher.
Please send any applications or enquiries to the address given.
Thank you.
L.Kalia
********************************************************************
STUDENTSHIP (PhD, 6k per annum, Psychology/Neural Networks)
Worcester College of Higher Education is funding a studentship (6000 pounds per
annum) for three years, to register for a PhD, fees paid.
AREA: Memory and Attention. PhD could involve either experimental work or
neural network simulations.
With: Dr.E.P. Fulcher from September 1994
To Apply:
Write (SAE) for further details and application form to:
Director of Personnel, Worcester College of Higher Education,
Henwick Grove, Worcester, WR2 6AJ.
QUOTE: RS 3 (1994)
CLOSING DATE: May 9th 1994.
********************************************************************
------------------------------
Subject: Auditory Modelling Software from Loughborough (LUTEar 1.5.0)
From: "L.P.OMard" <L.P.OMard@lut.ac.uk> (Lowel P. O'Mard)
Date: Fri, 15 Apr 1994 09:50:07 -0000
Dear All,
Please find below the README file for the latest version of the LUTEar
Core Routines Library (CRL). The UNIX and Macintosh (THINK C 5.0) platform
versions are now available (the MSDOS, Borland C, version will be ready by
Friday at the latest) via anonymous FTP from:-
suna.lut.ac.uk (131.231.16.2): /public/hulpo/lutear
Connect via FTP then login with user name "anonymous" and give your e-mail
address as the password. Download and thead the "INSTALL150" file from the
"/public/hulpo/lutear" directory (as also given above), then follow the
installation procedure for your platform. If you have any problems at all,
do not hesitate to get in contact with me.
Any comments, improvements, additions or corrections you may wish to
suggest are very welcome; it is only by direct feed-back from users that I
can ensure that the Core Routines Library is a delight to use, as well as
implementing state of the art auditory models.
...Lowel.
1. Introduction
As computer modelling of the auditory system increased in
complexity the need for common working tools became more pressing.
Such tools are necessary to allow the rapid dissemination of new computer
code, and to permit other members of the scientific community to replicate
and challenge published results.
The auditory models developed by the Speech and Hearing
Laboratory, at Loughborough University of Technology (UK.), have
received much attention, due principally to their simple form and the
many published papers in which the models are used to explain auditory
phenomena. The many requests for the computer code of the model
simulations led to the group releasing the LUTEar Core Routines Library
(CRL, version 1.0.0, October 1993) as a computational platform and set of
coding conventions which supports a modular approach to auditory system
modelling. The system is written in ANSI-C and works on a wide range of
operating systems. LUTEar has now been consolidated and much improved
in the latest release (version 1.5.0).
The CRL brings together established models, developed by the group,
and also contributed by other researchers in the field, which simulate
various stages in the auditory process. Since the first release, the LUTEar
CRL has been tested and used both at the originating laboratory and at
many other sites. It has been used as a tool for speech processing, speech
and voice analysis as well as in the investigation of auditory phenomena,
for which it was primarily constructed. This latest version of the CRL is a
product of the proving ground to which it was subjected, and we hope that
it will be as well received as was the first version.
Included with this release is a comprehensive series of test
programs. These programs were used to test the CRL routines; they
reproduce the behaviour of the respective published models included. The
programs also provide examples of how the CRL may be used in auditory
investigation programs. In addition the programs read data from
parameter files, and thus can be readily used to investigate further the
behaviour of the models included in the CRL.
The CRL routines have been subjected as much as possible to careful
and exhaustive testing. No system, however, is infallible so it is hoped
that, with the gentle admonitions of the library's users, any problems or
omissions will be quickly corrected. In addition it is expected that the
library will be augmented by further models as the scientific endeavour
continues. Many weeks have been required to get the manual into its
current form. It is not perfect, so gentle admonitions and suggested
changes/additions are invited.
1.1. CRL Features
The library has a modular structure which can be used to create
auditory investigation/application systems or incorporated in existing
code, as required. The library is intuitive in application, and has
comprehensive error reporting embedded in efficient code. All the
modules conform to a simple standard format. The design allows for
plugging and unplugging alternative models of the same component
auditory process for purposes of comparison. Ultimately the CRL is a
development based on the meld of experimental investigation methods and
the tenets of good software engineering practice. The following is a list of
the principal features of the CRL:-
o Modular Structure;
o Processing stage data can be handled by a single unit;
o Processing stage units can link to data from other stages;
o Multi-channel data is handled invisibly;
o Efficient algorithms are used throughout;
o Meaningful routine and variable names are used;
o All routines are prefixed by their module name;
o Comprehensive error handling incorporated in routines.
1.1.1 Main features new in version 1.5.0
o Improved manual: greater detail with over 65 figures and an index.
o Sound data file format reading/writing support;
o Connection management system (invisible to user);
o Modules can now read/print their own parameters;
o Generic programming introduced;
o New analysis routines, including FFT's;
o Binaural processing support;
o Non-linear basilar membrane filter model;*
o Stochastic inner hair cell model;*
o McGregor neural cell model;
o Dendrite filter model;
o Spike generation module (for Meddis86 IHC model output);
o New stimulus generation modules.
o Parameter files can have comment or blank lines;
o Direction of warnings and error messages to a specified file;
* These models are still in development, prior to publishing, but they have
been included for those who
may wish to look at them.
+-------------------------+-----------------------------------------------+
|Lowel P. O'Mard PhD. | /\ / \ Speech & Hearing |
|Dept. of Human Sciences, | /\/\ /\/ \/ /\ \ /\ Laboratory |
|University of Technology,|_/\/\/ /\ \/\/ /\ /\/ \ \/ /\/\_ /\___ |
|Loughborough, | \/\/ \/\/\/ \/ /\ \/\/ /\ / |
|Leics. LE11 3TU, U.K. | \ /\/\/\ /\/ \ /\/\/ \/ Director: |
|L.P.OMard@lut.ac.uk | \/ \/ \/ Prof. Ray Meddis |
+-------------------------+-----------------------------------------------+
------------------------------
Subject: Positions for PhD students
From: jb@informatik.uni-bonn.de
Date: Tue, 19 Apr 1994 18:13:59 +0100
!!!!!!!!!!!!!!!!!!!!!!!! A T T E N T I O N !!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!! A T T E N T I O N !!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!! A T T E N T I O N !!!!!!!!!!!!!!!!!!!!!!!!
Research Positions available in Image Analysis/Pattern Recognition at
the University of Bonn.
Two positions for PhD students/Postdocs are open at the Computer
Science Department of the University of Bonn starting Juli 1,
1994. One position is available to conduct research in image sequence
analysis for surveillance applications, the other is dedicated to
video compression applications.
(Salary at the research associate level)
Interested candidates should have a background in one of the following
research field:
* computer vision and image processing
* statistical pattern recognition
* neural networks and/or connectionists modeling
Applicants should send the Curriculum Vitae and a description of their
research interests to
Prof. J. Buhmann
Institut fuer Informatik III Tel.: +49 228 550 380
Universitaet Bonn Fax: +49 228 550 382
Roemerstr. 164 email: jb@informatik.uni-bonn.de
D-53117 Bonn jb@cs.bonn.edu
Fed. Rep. Germany
------------------------------
Subject: postdoc position in NN available
From: Joerg Kindermann <joerg@nathan.gmd.de>
Date: Wed, 20 Apr 1994 10:50:40 +0200
!!!!!!!!!!!!!!!!!!!!!!!! A T T E N T I O N !!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!! A T T E N T I O N !!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!! A T T E N T I O N !!!!!!!!!!!!!!!!!!!!!!
The research group "Adaptive Systems" of the German National Research
Center for Computer Science (GMD) in Sankt Augustin near Bonn has a
postdoctoral research position available. The group, headed by
Dr. Muehlenbein, consists of about 15 researchers working in the areas
of genetic algorithms, reflective statistics and robotics. Appointment
will be for two years (as of 1 July 1994 or later). It can possibly be
extended by another year.
Applicants should have a good background and strong research
interests in one or more of the following areas:
- statistical properties of neural networks and related learning
algorithms
- predictive reliability of statistical models
- adaptive algorithms
The successful candidate is expected to contribute actively to our
research projects in the area of reflective exploration (selective
sampling, model selection) with neural networks and related
statistical methods. Good programming skills in C++ or C are required.
Applicants should send their Curriculum Vitae and a description of
their research interests one of the addresses below. Applications
should be received by May 8, 1994.
Dr. Gerhard Paass Dr. Joerg Kindermann
paass@gmd.de kindermann@gmd.de
phone: +49 02241 142698 phone: +49 02241 142437
System Design Technology Institute
German National Research Center
for Computer Science (GMD)
Schloss Birlinghoven
D-53754 St. Augustin
Germany
------------------------------
Subject: new book
From: tibs@utstat.toronto.edu
Date: Sun, 17 Apr 1994 21:45:00 -0400
The following new book may be of interest to connectionists:
An Introduction to the Bootstrap- Brad Efron and Rob Tibshirani
This is the first general book written on the bootstrap and related topics
(Jackknife, cross-validation ...)
The purpose of this book is to present an overview of the bootstrap and
related methods for assessing statistical accuracy. The objectives are a)
to provide the reader with a working knowledge of bootstrap and related
methodologies, and b) serve as a resource text for researchers in the area.
The first 19 chapters are expository and are suitable for a one semester
course at the upper undergraduate or masters level. They require one
probability and one statistics course as a prerequisite. Each chapter has
numerous problems. We have written this part of the book so that it will
be accessible to non-specialists, particularly scientists who are seeking
to learn about these methods for possible use in their own work. The
remaining chapters are at a higher mathematical level, and together with
parts of Chapters 6-19, are suitable for a graduate level course in
statistics.
This book is aimed at statisticians, upper year undergraduate and graduate
students in statistics, and scientists, engineers and doctors who do
quantitative research.
Bradley Efron is the inventor of the bootstrap and is responsible for many
of the major research advances in the area. Robert Tibshirani was a student
of Dr. Efron's, has contributed to research in this area and is an active
researcher and author in the statistical community.
Ordering information:
An Introduction to the Bootstrap- Efron and Tibshirani ISBN 0-412-04231-2
Chapman and Hall
One Penn Plaza- 41st floor
New York, Ny. 10119
Phone 212 564-1060
Customer service FAX 212-268-9964
Toll free order FAX 1-800-248-4724
=============================================================
| Rob Tibshirani
To every man is given the key to | Dept. of Preventive
the gates of heaven; | Medicine and Biostatistics
the same key opens the gates of hell. | McMurrich Bldg.
| University of Toronto
Buddhist proverb | Toronto, Canada M5S 1A8
Phone: 1-416-978-4642 (biostats) Email: tibs@utstat.toronto.edu
416-978-0673 (stats)
FAX: 1-416-978-8299
------------------------------
Subject: Contents of NETWORK, Volume 5, Number 1.
From: David Willshaw <david@cns.edinburgh.ac.uk>
Date: Wed, 20 Apr 1994 11:48:39 -0000
Network: Computation in Neural Systems
Volume 5 Number 1 February 1994
----------------------------------------
PAPERS
1 Hebbian learning is jointly controlled by electrotonic and input
structure
K Y Tsai, N T Carnevale and T H Brown
21 Efficient mapping from neuroanatomical to electrotonic space
K Y Tsai, N T Carnevale, B J Claiborne and T H Brown
47 Regulating the nonlinear dynamics of olfactory cortex
Xingbao Wu and H Liljenstrom
61 Spontaneous symmetry breaking and the formation of columnar
structures in the primary visual cortex
K Yamagishi
75 Using generalized principal component analysis to achieve
associative memory in a Hopfield net
S Coombes and J G Taylor
89 Learning temporal sequences by excitatory synaptic changes only
Y Metzger and D Lehmann
101 Hierarchical neural networks for time-series analysis and
control
T Frohlinghaus, A Weichert and P Rujan
117 ABSTRACTS SECTION
119 BOOK REVIEWS
119 An introduction to the modeling of neural networks
P Peretto
120 Computational learning theory
M Anthony and N Biggs
NETWORK welcomes research Papers and Letters where the findings have
demonstrable relevance across traditional disciplinary boundaries.
Research Papers can be of any length, if that length can be justified
by content. Rarely, however, is it expected that a length in excess
of 10,000 words will be justified. 2,500 words is the expected limit
for research Letters.
Articles can be published from authors' TeX source codes. Macros can
be supplied to produce papers in the form suitable for refereeing and
for IOP house style. For more details contact the Editorial Services
Manager at IOP Publishing, Techno House, Redcliffe Way, Bristol BS1
6NX, UK.
Telephone: (+44) 0272 297481
Fax: (+44) 0272 294318
Telex: 449149 INSTP G
Email Janet: net@uk.co.ioppublishing
Subscription Information
Frequency: quarterly
Subscription rates: Institution 192.00 pounds (US$376.00)
Individual (UK) 32.00 pounds
(Overseas) 35.00 pounds (US$75.00)
A microfiche edition is also available
------------------------------
Subject: book listing
From: benirsch@cshl.org (Ingrid Benirschke)
Date: Thu, 21 Apr 1994 16:39:51 -0400
[[ Editor's Note: Long-time readers know about my non-commercial policy.
Even though the price of this volume is a bit steep, I thought the
announcement was within the bounds of reason. Further, the volume may be
of interest to people who wish to apply connectionist models to this
field. -PM ]]
MOLECULAR PROBES OF THE NERVOUS SYSTEM
Volume 1: Selected Methods for Antibody and Nucleic Acid Probes
By Susan Hockfield, Yale University School of Medicine
Steve Carlson, University of Washington
Chris Evans, University of California, Los Angeles
Pat Levitt, Medical College of Pennsylvania
John Pintar, Columbia University College of Physicians & Surgeons
Laura Silberstein, San Jose State University
Over the past ten years, new techniques have transformed most branches
of neuroscience. The most powerful of them have been adapted from
cell and molecular biology and immunology to permit analysis of nerve
cell structure and function at a level of detail never before possible. With
this volume, CSHL Press introduces a new series of manuals designed
specifically for neuroscientists who wish to acquire the skills of molecular
analysis. A laboratory course taught each year at Cold Spring Harbor
Laboratory demonstrates the potential and pitfalls of antibody and nucleic
acid probes as biochemical and anatomical reagents. This first volume in
the series, designed for ease of use at the lab bench, is a distillation by the
instructors of much of the material covered in several years of the course.
As an aid to understanding the strategy and rationale of the protocols, brief
supplementary explanations are included. The manual will be a valuable
resource for established investigators and for anyone working with nervous
system tissue for the first time.
1993, 656 pp., illus., color plates, appendices, index
Plastic comb binding $85 ISBN 0-87969-351-7
Cloth $170 ISBN 0-87969-372-X
For further information, or to place an order, you can call Cold Spring Harbor
Laboratory Press at 1-800-843-4388, fax 516-349-1946, or e-mail "benirsch@
cshl.org"
Ingrid Benirschke
Marketing Manager
Tel: 516-367-8470
Fax: 516-367-8532
e-mail: benirsch@cshl.org
------------------------------
End of Neuron Digest [Volume 13 Issue 27]
*****************************************
