KDD Nuggets Index


To KD Mine: main site for Data Mining and Knowledge Discovery.
To subscribe to KDD Nuggets, email to kdd-request
Past Issues: 1996 Nuggets, 1995 Nuggets, 1994 Nuggets, 1993 Nuggets


Data Mining and Knowledge Discovery Nuggets 96:21, e-mailed 96-06-28

News:
* T. Anand, updated list of demos at KDD-96
* G. Williams, Australian Data Mining Group and Reports
http://www.dit.csiro.au/~gjw/dataminer.html
* D. Cohen, CART from California Statistical Software ?
* R. Golan, Financial Data Mining with Rough Sets
* GPS, ComputerWorld on going beyond Data Warehousing,
http://www.computerworld.com/search/AT-html/9606/960624SL26patty.html
Publications:
* GPS, AI Magazine Summer 1996 on Data Mining a Sky Survey
Meetings:
* G. Widmer, ECML-97: Call for Papers
http://is.vse.cz/ecml97/home.html
* J. Ignacio, Workshop on Machine Learning and more in Madrid,
JULY 10-12, 1996

--
Data Mining and Knowledge Discovery community,
focusing on the latest research and applications.

Contributions are most welcome and should be emailed,
with a DESCRIPTIVE subject line (and a URL, when available) to (kdd@gte.com).
E-mail add/delete requests to (kdd-request@gte.com).

Nuggets frequency is approximately weekly.
Back issues of Nuggets, a catalog of S*i*ftware (data mining tools),
and a wealth of other information on Data Mining and Knowledge Discovery
is available at Knowledge Discovery Mine site, URL http://info.gte.com/~kdd.

-- Gregory Piatetsky-Shapiro (moderator)

********************* Official disclaimer ***********************************
* All opinions expressed herein are those of the writers (or the moderator) *
* and not necessarily of their respective employers (or GTE Laboratories) *
*****************************************************************************

~~~~~~~~~~~~ Quotable Quote ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We are always given the finite amounts of data ...
and rarely do we reach asymptotia.
Ron Kohavi, 1996 (in Data Mining using MLC++ paper,
ftp://starry.stanford.edu/pub/ronnyk/mlc96.ps.Z

Previous  1 Next   Top
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Tue, 25 Jun 1996 10:23:39 -0400
From: tanand@winhitc.atlantaga.ncr.com (Tej Anand)
Subject: RE: KDD96 demos -- updated list

DBMiner: A System for Mining Knowledge in Large Relational Databases
Jiawei Han, Yongjian Fu, Wei Wang, Jenny Chiang, Wan Gong, Krzysztof Kope=
rski,
Deyi Li, Yijun Lu, Amynmohamed Rajan, Nebojsa Stefanovic, Betty Xia, Osma=
r R.
Zaiane.
School of Computing Science, Simon Fraser University, British Columbia, C=
anada
V5A 1S6
(han@cs.sfu.ca)

Kepler: Extensibility in Data Mining Systems
Stefan Wrobel, Dietrich Wettschereck, Edgar Sommer, Werner Emde GMD, FIT.=
KI,
Schloss Birlinghoven, 53754 Sankt Augustin, Germany
(Stefan.Wrobel@gmd.de)

WebFind: Mining external sources to guide WWW discovery.
Alvaro E. Monge and Charles P. Elkan
Department of Computer Science and Engineering University of California, =
San
Diego, La Jolla, CA 92093-0114 =

{amonge,elkan}@cs.ucsd.edu

MM -- Mining with Maps
Raymond Ng
Department of Computer Science, University of British Columbia
(rng@cs.ubc.ca)

Decisionhouse
Nick Radcliffe
Quadstone Ltd., UK
(njr@quadstone.co.uk)

STARC - A New Data Mining Tool
Damir Gainanow, Andre Matweew, Michael Thess DATA-Center Ltd., Ekaterinbu=
rg,
RUSSIA
Scholz & Thess Software GbR, Chemnitz, GERMANY
(michael.thess@Mathematik.TU-Chemnitz.DE)

D-SIDE: A probabilistic DeciSIon enDorsement Environment P.Kontkanen,
P.Myllym=E4ki and H.Tirri
Complex Systems Computation Group, Department of Computer Science Univers=
ity of
Helsinki, Finland
(myllymak@cs.Helsinki.FI)

Optimization Related Data Mining using the PATTERN System H. Bodek, R. L.=

Grossman, D. Northcutt, H. V. Poor Magnify, Inc.
Princeton University
(rlg@opr.com)

Clementine Data Mining System
Colin Shearer
Data Mining Division, Integral Solutrions Ltd
(colin@isl.co.uk)

MineSet
Steven Reiss, Mario Schkolnick =

Data Mining and Visualization Group, Silicon Graphics Computer Systems
(sreiss@powerplay.engr.sgi.com)

FACT: Finding Associations in Collections of Text
Ronen Feldman, Haym Hirsh
Bar-Ilan Univesity
Rutgers University.
(feldman@cs.biu.ac.il) =


DataMine - An Integrated Knowledge Discovery Environment
Tomasz Imielinski and Aashu Virmani
Rutgers University
(avirmani@paul.rutgers.edu)

Data Surveyor
M.Holsheimer, F.Kwakkel, D.Kwakkel, P. Boncs.
Data Distilleries, Kruislaan 419, 1098 VA Amsterdam,
The Netherlands.
Ad Feelders (ad@ddi.nl)

Management Discovery Tool
Ken O'Flaherty
NCR Corporation
(ken.o'flaherty@sandiegoca.ncr.com)

WEBSOM - Interactive Exploration of Document Collections
Krista Lagus, Timo Honkela, Samuel Kaski and Teuvo Kohonen
Neural Networks Research, Centre of Helsinki University of Technology
krista@nucleus.hut.fi

IBM Data Mining Tools
Julio Ortega, Kamal Ali, Stefanos Manganaris, George John
IBM Almaden Research Center
(julio@almaden.ibm.com)

Geomarketing decision support system
Cyril Way, Hugues Marty, Thierry Marie Victoire
Isoft
(erict@isoftfr.isoft.fr)

Ac2: Advanced decision tree based Data Mining
Cyril Way, Hugues Marty, Thierry Marie Victoire
ISoft
(erict@isoftfr.isoft.fr)

Previous  2 Next   Top
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Tue, 25 Jun 1996 13:12:54 +1000 (EST)
From: Graham Williams (Graham.Williams@cbr.dit.csiro.au)
Subject: Data Mining Group and Reports

CSIRO is the Australian Government's research organisation. The
CSIRO Division of Information Technology and Division of Mathematics
and Statistics have a long history in Databases, Artificial
Intelligence, and Exploratory Data Analysis. Researchers in Machine
Learning, Classification, Databases, and Statistics have come together
to cooperate on the CSRIO DIT Data Mining Portfolio. The project is
affiliated with the Australian National University's Cooperative
Research Centre for Advanced Computational Systems. The group has
been active over the past year with a number of partners from the
insurance and health industries.

The group's home page can be found at:

http://www.dit.csiro.au/~gjw/dataminer.html

A number of relevant technical reports are available.

--
Dr Graham.Williams@cbr.dit.csiro.au ,--_| Tel: (+61 6) 216 7042
Senior Research Scientist, Data Mining / Fax: (+61 6) 216 7112
CSIRO Div of Info Tech, Aust Nat Univ _.--_*/ GPO Box 664 Canberra
http://www.dit.csiro.au/~gjw v ACT 2601 Australia



Previous  3 Next   Top
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Tue, 25 Jun 1996 09:05:36 -0400
From: Dawn Cohen (dcohen@cs.pitt.edu)
Subject: CART

...I am trying to get a copy of the CART Classification and
Regression Trees software from California Statistical Software. I am having
a very hard time locating them, so if you know or anyone on the mailing list
could help me out on this, I'd appreciate it greatly.

--Dawn Cohen

[I know that CART method is included in IND package
see http://info.gte.com/~kdd/siftware.html#IND or
http://cognac.cosmic.uga.edu/abstracts/arc-13188.html,
but is there a separate standalone CART module ? if yes, I would
appreciate if someone sends the latest info about it to kdd@gte.com]


Previous  4 Next   Top
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[I include a financial application description forwarded by
Robert Golan -- more in next issue. GPS]
Date: 13 May 1996 12:59:24 +0500
From: 'Robert Golan' (Robert_Golan@mail.tcpl.ca)
Subject: Financial Data Mining From CIFEr '96 conference

USING ROUGH SETS FOR MARKET =
TIMING

My name is Murray A. Ruggiero, Jr. I am the former Vice-President of =
Promised Land Technologies, Inc. and the inventor of a patented method for =
embedding a neural network into a spreadsheet. I have a bachelor's degree =
in physics/ astronomy and computer science. My research in using =
artificial intelligence to develop market timing applications dates back =
to 1988. I was also featured in the November 2, 1992, Business Week =
article - 'The New Rocket Science' as one of the experts interviewed =
about using neural networks in finance.



My company, Ruggiero Associates, was started in January, 1993. Ruggiero =
Associates specializes in developing market timing applications using =
state-of-the-art technologies. Ruggiero Associates services include =
training traders to use these new technologies for developing their own =
trading systems. Ruggiero Associates has a broad range of clients =
ranging from high profile individual traders such as Larry Williams to =
institutional money managers.



I am Contributing Editor at Futures Magazine and write a monthly column =
on using advanced technologies such as neural networks and rough sets in =
trading.



Ruggiero Associates has developed many proprietary =
trading systems for our clients using rough sets. Developing trading =
systems has many challenges. Some of these are as follows:



1) Defining a target which is useful for trading

2) Developing preprocessing which is predictive of that =
target.

3) Avoiding overfitting of the data

4) Selecting rules that have the best probability of =
continuing to work into the future.

5) Combining domain expertise with machine generated rules to =
improve tradability.





Let's now discuss each of these challenges in more detail using the =
S&P500 as an example.



Our first challenge was to define a target that was useful for trading =
the S&P500 without holding any position overnight. We call this =
day-trading. Next, we had to define this target using daily data and not =
intraday bars. We solved this problem by selecting a simple target that =
was defined as the sign of the close minus the open.



Our next challenge was to define inputs that were predictive =
of our desired output. The S&P500 has a positive correlation to prices =
of both long and short term interest rates. For this reason, we chose =
S&P500, 30 year T-Bond and Eurodollar futures. We used data from 4/21/82 =
to 12/31/92 to develop our model. Our test data was 1/1/93 to 5/31/94 and =
our blind set was 6/1/94 to 1/1/95. Besides these price series, we also =
used day of week. Day of week has a statistically significant effect on =
the S&P500. Let's look at the effect of day of week on the S&P500.



We analyzed the day of week effect from 4/21/82 to 11/24/95. During this =
period, buy and hold produced 484.70 points. Since we want to see market =
tendencies, we have not deducted any slippage and commissions.



Day of Week Net Change Ave Change %Buy and Hold

Monday 264.10 .387 54.4%

Tuesday 6.05 .009 1.2%

Wednesday 192.80 .275 39.8%

Thursday 52.50 .075 10.8%

Friday -115.39 -.167 -23.8%



Figure 1

As you can see in Figure 1, just buying on Monday and Wednesday produces =
over 94% of buy and hold while only being in the market 40% of the time. =
You can also see that Friday has a significant downward bias.



After selecting our raw data we need to preprocess it. We =
analyzed each of these price series to see how to preprocess them so they =
would be predictive of our desired outputs. We used a series of simple =
percent changes at 1,2,5,7,10, 15, 25 days. We also used each price =
series by comparing prices relative to a moving average and sampled them =
using the same sampling as the percent changes. The moving average =
lengths were selected by selecting them based on their performance in a =
simple trading system.



We then preprocessed both the percent changes and =
price relative to a moving average by only using the sign of these =
numbers. Besides this price series information for each series, we also =
added day of week. Once again, our output was the sign of the close minus =
the open.



We then analyzed this data using Reduct Systems Data Logic/R. We used a =
precision of 60% and a roughness ranging between .5 and .9. One of the =
reasons we selected such low values for roughness is that we wanted to =
avoid overfitting. This is also why we selected the 60% probability level =
on the rules outputted by rough sets. If we required 70-80% accuracy, the =
model would have been less robust.



We analyzed the results of the rules generated by rough sets =
in order to judge the probability that they would continue to work into =
the future. We then ranked the rules. The less conditions a rule =
contained, the higher its ranking. Next, we took the ratio of the number =
of conditions divided by the percentage of supporting cases for each rule. =
We ranked the rule higher as the ratio of cases to conditions increased. =
We also selected rules that were more than 65% accurate at discriminating =
our output and had at least 10% of the cases in that database support =
them. Next, we examined where each case occurred in the data set and =
scored each rule higher based on a uniform distribution of occurrences.



We then analyzed each rule that met our criteria using domain expertise =
in order to select rules that had the highest probability of continuing to =
perform well in the future. These rules were then translated these rules =
into Omega Research's TradeStation's EasyLanguage and tested. We tested =
them on three data sets. The first was the period used in developing the =
rules. Next, we used the test set. After finishing our final selection =
process, we will test the rules on our blind data set.



We first tested each rule we have selected on the =
development set in order to judge its tradability. For example, many rules =
had a high winning percentage but the average trade was too small to =
trade the rule successfully with slippage and commissions. We then =
analyzed rules to find ones that were tradable during the development =
period and the test set. We only selected rules that had similar =
performance during both the development and test period. If the rules do =
not have similar performance, but the performance is good across both =
sets, you should then try to find factors between these periods that can =
account for the differences.



Using many of the rules, we found that some of them worked =
well during the development period and badly during the test period. We =
also found that other rules worked well during the development period and =
then worked even better during the test period. For example, some rules =
produced $300.00 a trade during the development period and $700.00 during =
the test period. Both of these results are a problem. For a system to =
continue to work into the future it needs to have similar performance over =
both data sets.



When developing a system for Larry Williams, I discovered that some of =
the trading rules worked much better out of sample than they did during =
the development set. These changes in the distribution of trades between =
the development and out of sample sets concerned me. When I analyzed the =
results closer, I discovered that rules that worked well from October =
1989 to October 1992, did not work well from October of 1992 to November =
of 1994. I later found that they started to work well again in =
mid-November of 1994. The reverse was also true. Many rules that worked =
well form October of 1992 to November of 1994 did not work well during the =
surrounding periods.

I needed to discover what causes these changes. The first thing that my =
analysis showed was that these boundaries occurred around major =
corrections in the market. I then analyzed the changes in correlation =
around these corrections and found that the S&P500 and T-Bonds usually =
have a breakdown in correlation around market corrections. An even more =
amazing fact is that the long term correlation between stocks and T-Bonds =
alternates between strong and weak based on major market events.



Period Pearson's correlation

October 13 1989 to October 2 1992 .760

October 3 1992 to November 17 1994 .269

November 18 1994 to August 16 1995 .944

July 19 1995 to August 16 1995 .355 (Not enough days to =
confirm)



We found that we could classify how well a given rule would work based on =
long term intermarket correlations. We then tested how shorter term =
moving correlation affected rule performance. We found that using a 10 to =
30 day correlation allowed us to discriminate between rules that had a =
high probability of working at any given point in time. We then added the =
correlation between the S&P500 and T-Bonds and the S&P500 and Eurodollars =
to our rules. We found that some rules work better when the correlations =
are above a given level and some when it is below.



One of these models that I developed for the S&P500 last year =
for a client has had 18 trades over the period 4/19/95 and 3/1/96. During =
this period, it has made $10,125.00 on these 18 trades or $562.50 a trade =
without holding an overnight position. It also won thirteen out the 17 =
trades. This system is based on four different rules produced by rough =
sets. One of the rules outperformed the other rules in this system. This =
one rule produced nine trades and eight winners during this same period. =
These trades produced an $8,100.00 net profit and an average trade of =
over $900.00.



These results show the power of rough sets as a tool for =
building trading systems. Combined rough sets with domain expertise can =
produce results that previously traders only dreamed of.


Previous  5 Next   Top
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Wed, 26 Jun 1996 15:49:05 -0400
From: Gregory Piatetsky-Shapiro (gps@gte.com)
Subject: Customer care systems: The next generation 6/24/96
X-Url: http://www.computerworld.com/search/AT-html/9606/960624SL26patty.html


Customer care systems: The next generation



Patricia Seybold

06/24/96




Recently I attended a meeting of information systems
professionals in the telecommunications industry. What a tough job.



The industry is undergoing intense competition from upstarts who
aren't encumbered by legacy systems and legacy policies. The systems
are large, complex and hard to change. The interfaces between the
order-taking systems and the operational systems are hard-coded and
based on 20-year-old industry specifications.



Meanwhile, the business executives are tearing out their hair
because they need much tighter relationships with their customers the
customers whose data is locked in to these hard-coded applications
before those customers are stolen away by the competition. The
business executives also need to be able to design and launch new
products and services in weeks, not months.



Does this sound familiar? Banking, energy, utility,
pharmaceutical and other industries are facing similar scenarios.



The simple answer, you might say, is to build a data warehouse to
pull all the customer-related information in to one logical
place. Then, create a SWAT team with rapid application development
(RAD) tools to prototype, design and roll out applications for new
products and services.



Wrong! Eventually, you'll probably need to do both of those
things, but they won't solve the main problem or let you get the
well-deserved rest and recreation you crave.



What you need is a new platform for integrated customer
interactions. Replace your outmoded order-taking systems with
state-of-the-art customer-interaction applications. Customers should
be able to sign up for your products and services by calling your
toll-free telephone number (as they probably do now) or by following
the prompts on a voice-response system. Or, they should be able to
jump on the Internet to request information, order services, call up
their bill and pay it, review their profile of services and change
some of them.



Furthermore, you want every employee from the salespeople and
technicians in the field to the vice president of marketing to have
access to an integrated picture of the firm's interactions with your
customers. The applications they use to do their jobs should be
integrated with the customer-interaction platform.



This is quite different from a conventional data warehouse. A
data warehouse consolidates information stored in legacy applications,
refines it and puts it in understandable form for analysis. What I'm
talking about here is a dynamic suite of applications with real-time
information. These are the systems you use to serve customers, to
cement and strengthen customer loyalty and to discern patterns so that
you can create popular new programs and services.



The system should provide a 360-degree view of customers and
their interactions with your firm. The integrated suite of
applications will handle everything from billing to field sales, from
help desk support to enabling customers to help themselves to
information and services.



You can start with one of the many off-the-shelf packages for
integrated customer care and tailor it to your business. Or, you can
build it yourself, knocking off one application at a time.



Yes, you'll want to have a data warehouse full of relevant
information about your customer accounts. Yes, you'll want a RAD SWAT
team to develop and deploy applications for new product rollouts. But
do all of this in the context of an integrated customer-interaction
platform. That's how you become a hero and get to enjoy that
well-deserved vacation.




Previous  6 Next   Top
>~~~Publications:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Fri, 28 Jun 1996 08:40:57 -0400
From: gps@gte.com (Gregory Piatetsky-Shapiro)
Subject: AI Magazine Summer 1996 on Data Mining a Sky Survey

The latest (Summer 1996) AI magazine has an article by Fayyad,
Djorgowski, and Weir, describing the SKICAT system -- one of the best
known successes of Data Mining in the scientific field.

SKICAT has been described in previous KDD and AAAI meetings, and
this article provides a good overview and sums the latest results
on unsupervised learning and new scientific discoveries.


Previous  7 Next   Top
>~~~Meetings:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Thu, 27 Jun 1996 12:19:16 +0200
From: Gerhard Widmer (gerhard@ai.univie.ac.at)
Subject: ECML-97: Call for Papers http://is.vse.cz/ecml97/home.html

ECML-97

9th EUROPEAN CONFERENCE ON MACHINE LEARNING

23-26 April 1997, Prague, Czech Republic

Call for Papers

-----------------------------------------------------------------------
Up-to-date information on the conference can be found at
http://is.vse.cz/ecml97/home.html
_______________________________________________________________________


GENERAL INFORMATION:

The 9th European Conference on Machine Learning (ECML-97)
will be held in Prague, Czech Republic, during April 23-26, 1997,
with informal workshops on April 26.
The goal of ECML is to be a forum for the discussion of research in
and applications of all forms of machine learning. Although the
emphasis is on scientific advances in machine learning, ECML
also requests papers on applications to practical problems or
to other sciences, provided that general implications of the
application are pointed out.

One of the explicit goals of ECML-97 is to widen the audience and
to strengthen relations between machine learning and other fields
such as statistics, cognitive science, knowledge acquisition,
linguistics, databases, etc.


PROGRAM:

The scientific program (April 23-25) will include invited
talks, presentations of accepted papers, poster and demo
sessions, as well as summary and commenting sessions on current
and upcoming issues in machine learning.
Saturday, April 26, will be devoted to informal workshops, for
which a separate call for proposals will be published
(contact Maarten van Someren (maarten@swi.psy.uva.nl) for details).


RELEVANT RESEARCH AREAS:

Submissions are invited in all areas of Machine Learning,
including, but not limited to:

abduction analogy
applications of machine learning artificial neural networks
case-based learning computational learning theory
evolutionary computation inductive learning
inductive logic programming knowledge base refinement
knowledge discovery in databases knowledge-intensive learning
language learning learning and problem solving
models of human learning multi-agent learning
multistrategy learning reinforcement learning
revision and restructuring robot learning
scientific discovery statistical approaches


PROGRAM CHAIRS:

Maarten van Someren (University of Amsterdam) and
Gerhard Widmer (University of Vienna and Austrian Research
Institute for Artificial Intelligence, Vienna).


LOCAL CHAIR:

Radim Jirousek (University of Economics, Prague).


PROGRAM COMMITTEE:

D. Aha (USA) F. Bergadano (Italy)
I. Bratko (Slovenia) P. Brazdil (Portugal)
K. De Jong (USA) L. De Raedt (Belgium)
S. Dzeroski (Slovenia) W. Emde (Germany)
Y. Kodratoff (France) N. Lavrac (Slovenia)
R. Lopez de Mantaras (Spain) H. Mannila (Finland)
S. Matwin (Canada) K. Morik (Germany)
G. Nakhaeizadeh (Germany) C. Rouveirol (France)
L. Saitta (Italy) J. Schmidhuber (Switzerland)
D. Sleeman (UK) P. Vitanyi (Netherlands)
S. Wrobel (Germany)


SUBMISSION OF PAPERS:

Two kinds of submissions are solicited: full papers describing
substantial completed research or applications, and poster
papers reporting on work in progress. Submissions must be
clearly marked as one of these two kinds.
The programme committee may decide to move accepted
contributions from the full paper to the poster category.
Full papers will be presented at plenary sessions and will
appear in the conference proceedings, poster papers will be
published in a technical report.

The size limit for submissions is 12 pages for full papers,
5 pages for poster papers (excluding title page and bibliography,
but including all tables and figures).
Submissions exceeding this limit will not be reviewed!

The conference proceedings will be published by Springer Verlag
as part of the 'Lecture Notes in AI (LNAI)' series.
Submitted papers should preferably be formatted according to
the LNAI guidelines (LaTeX style files are available at
http://is.vse.cz/ecml97/home.html or by sending an e-mail to
gerhard@ai.univie.ac.at). The publishers are also considering
to make the proceedings available electronically, before the
conference.

A separate title page must contain the title of the paper,
the names and addresses of all authors, up to three keywords,
and an abstract of max. 200 words. The full address, including
phone, fax and e-mail, must be given for the first author
(or the contact person).

The following items must be submitted by October 21, 1996:
Four (4) hard copies of the paper, an electronic version
(uuencoded, compressed PostScript) of the paper, and an
electronic version of the titlepage only (plain ASCII).
Send submissions, enquiries, etc. to:

Gerhard Widmer (ECML-97)
Austrian Research Institute for Artificial Intelligence,
Schottengasse 3, A-1010 Vienna, Austria
e-mail: gerhard@ai.univie.ac.at

Papers will be evaluated with respect to relevance, technical
soundness, significance, originality, and clarity. Papers
reporting on real-world applications will be evaluated
according to special criteria.
A copy of the review form, which specifies the criteria to
be used in the reviewing process, can be obtained electronically
from http://is.vse.cz/ecml97/home.html.


REGISTRATION AND FURTHER INFORMATION:

For information about paper submission and program, contact
the program chairs. For information about local arrangements,
registration forms, etc. contact the local organizers at
actionm@cuni.cz or check the ECML-97 WWW page.


IMPORTANT DATES:

Submission deadline: 21 October 1996
Notification of acceptance: 10 January 1997
Camera ready copy: 31 January 1997
Conference: 23-26 April 1997


IMPORTANT ADDRESSES:

Submission of papers to:
Gerhard Widmer (ECML-97)
Austrian Research Institute for Artificial Intelligence,
Schottengasse 3, A-1010 Vienna, Austria
e-mail: gerhard@ai.univie.ac.at

WWW site with conference and registration information,
LaTeX style files, sample review form, etc.:
http://is.vse.cz/ecml97/home.html

E-mail address of local conference organization:
actionm@cuni.cz


------------------------------------------------------------------------
Gerhard Widmer URL: http://www.ai.univie.ac.at/~gerhard
Austrian Research Institute e-mail: gerhard@ai.univie.ac.at
for Artificial Intelligence
Schottengasse 3 tel: +43-1-53532810
A-1010 Vienna, Austria fax: +43-1-5320652
------------------------------------------------------------------------


Previous  8 Next   Top
>~~~Meetings:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From: JIgnacio@grial.uc3m.es
Date: Thu, 27 Jun 1996 17:08:01 +0200


REGISTRATION FORM FOR FIRST INTERNATIONAL WORKSHOP ON MACHINE LEARNING,
FORECASTING, AND OPTIMIZATION 1996
WORKSHOP TO BE HELD ON JULY 10-12, 1996 AT UNIVERSIDAD CARLOS III DE MADRID


Tutorials

Tutorials will be in Spanish. Check off one tutorial for each
period of time. Tutorials will be available in case that at least 10 persons
register for that tutorial.

Wednesday, July 10, 1996

9:00 AM - 11:00 AM
__ Descubrimiento de relaciones en grandes Bases de Datos: Aurora Perez
(Universidad Politecnica de Madrid) and Angela Ribeiro (Instituto de
Automatica Industrial, CSIC)
__ Prediccion dinamica: series temporales: David Rios
(Universidad Politecnica de Madrid)

11:30 AM - 13:30 PM
__ Aprendizaje inductivo aplicado a la medicina: Cesar Montes
(Universidad Politecnica de Madrid)
__ Tecnicas de Inteligencia Artificial aplicadas a las finanzas: Ignacio
Olmeda (Universidad de Alcala de Henares)

15:00 PM - 17:00 PM
__ Analisis estadistico de datos: Jacinto Gonzalez Pachon
(Universidad Politecnica de Madrid)
__ Programacion Genetica en problemas de control: Javier Segovia (Universidad
Politecnica de Madrid)


Preliminary Workshop Schedule

Thursday, July 11, 1996
-----------------------

9:30 Registration
10:00 General presentation
10:10 Invited Talk: Manuela Veloso (Carnegie Mellon University)
11:30 Coffee Break
11:45 Session on Mathematical and Integrated Approaches
- 'Nonparametric Estimation of Fully Nonlinear Models for Assets
Returns', Ignacio Olmeda and Eugenio Fernandez
- 'Representation Changes in Combinatorial Problems: Pigeonhole
Principle versus Integer Programming Relaxation', Yury V. Smirnov
and Manuela M. Veloso
- 'Integrating Reasoning Information to Domain Knowledge in the Neural
Learning Process', Saida Benlarbi and Kacem Zeroual
- 'Parameter Optimization in ART2 Neural Network, using Genetic
Algorithms', Enrique Muro
1:15 Lunch
2:15 Invited Talk: Esther Ruiz (Universidad Carlos III de Madrid)
3:30 Session on Genetic Algorithms and Programming Approaches
- 'Automatic Generation of Turing Machines by a Genetic Approach',
Julio Tanomaru and Akio Azuma
- 'GAGS, a Flexible Object Oriented Library for Evolutionary
Computation', J.J. Merelo and A. Prieto
- 'An Application of Genetic Algorithms and Heuristic Techniques in
Scheduling', Celia Gutierrez, Jose M. Lazaro and Joseba Zubia
- 'Classifiers Systems for Learning Reactions in Robotic Systems',
Araceli Sanchis, Jose M. Molina and Pedro Isasi
- 'A Comparison of Forecast Accuracy between Genetic Programming and
other Forecastors: A Loss-Differential Approach', Shu-Heng Chen and
Chia-Hsuan Yeh



Friday, July 12, 1996
---------------------

10:00 Invited Talk: Juan J. Merelo (Universidad de Granada)
11:30 Coffee Break
11:45 Session on Neural Network Approaches
- 'Factor Analysis in Social Science: An Artificial Neural Network
Perspective', Rafael Calvo
- 'Neural Network Forecast of Intraday Futures and Cash Returns',
Pedro Isasi, Ignacio Olmeda, Eugenio Fernandez and Camino Fernandez
- 'Self-Organizing Feature Maps for Location and Scheduling',
S. Lozano, F. Guerrero, J. Larra~neta and L. Onieva
- 'A Neural Network Hierarchical Model for Speech Recognition based
on Biological Plausability', J.M. Ferrandez, D. del Valle,
V. Rodellar and P. Gomez
1:15 Lunch
2:15 Invited Talk: Alicia Perez (Boston College)
3:30 Session on Symbolic Machine Learning Approaches
- 'Statistical Variable Interaction: Focusing Multiobjective
Optimization in Machine Learning', Eduardo Perez and Larry Rendell
- 'A Multi-Agent Model for Decision Making and Learning',
Jose I. Giraldez and Daniel Borrajo
- 'An Approximation to Generic Knowledge Discovery in Database
Systems', Aurora Perez and Angela Ribeiro
- 'Learning to Forecast by Explaining the Consequences of Actions',
Tristan Cazenave
- 'Basic Computational Processes in Machine Learning',
Jesus G. Boticario and Jose Mira
5:15 Panel



Workshop Fees

Workshop fees are:
Paid before June 30 Paid after June 30
------------------- ------------------
Regular rate 25.000 pts. 35.000 pts.
Speakers rate 15.000 pts. 25.000 pts.
Students rate 5.000 pts. 10.000 pts.
Carlos III members 2.000 pts. 5.000 pts.
(students/teachers/staff)

The workshop fee includes a copy of the proceedings, coffee breaks, and
Thursday and Friday lunches. Students must send legible proof of full-time
student status.


Tutorial fees are:
Paid before June 30 Paid after June 30
------------------- ------------------
Regular rate 30.000 pts. 40.000 pts.
University rate 15.000 pts. 20.000 pts.
Students rate 2.500 pts. 5.000 pts.
Carlos III members 2.000 pts. 3.000 pts.
(students/teachers/staff)

The tutorials fee includes the assistance to three non-parallel tutorials,
documentation, and coffee breaks.


Registration

Please, send the completed application form to:
- by email: dborrajo@grial.uc3m.es, or isasi@gaia.uc3m.es
- by airmail:
MALFO96,
Attention D. Borrajo/P. Isasi
Universidad Carlos III de Madrid
c/ Butarque, 15
28911 Leganes, Madrid. Spain



Registration Form

Last name:__________________________________________________________________

First name:_________________________________________________________________

Title:______________________________________________________________________

Affiliation:________________________________________________________________

Address:____________________________________________________________________

____________________________________________________________________________

____________________________________________________________________________

Phone (include country and area code):______________________________________

Fax (include country and area code):________________________________________

E-mail:_____________________________________________________________________


Workshop fee:_________

Tutorial fee:_________
(please indicate which three tutorials do you wish to attend)



Total amount:_________


Send a check made payable to 'Universidad Carlos III de Madrid' or
electronic funds transfer to:
Account holder name: Universidad Carlos III de Madrid
Bank: Caja de Madrid
Branch address: c/ Juan de la Cierva s/n, Getafe
Bank code: 2038
Branch number: 2452
Account number: 6000085134
Control Digit: 05

In the case of a transfer, please indicate that it is code 396
(Inscripciones de Malfo96) and send us a copy of the receipt.
No refunds will be made; however, we will transfer your
registration to a person you designate upon notification.

Signature:__________________________________________________________________


Previous  9 Next   Top
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~