If you know what you're doing, use MATLAB for prototyping your
algorithms....?

Scott
PTIV Electronics with AI
University of Southampton


--- In artificialintelligencegroup@yahoogroups.com, Sarvi Shanmugham
<sarvi@c...> wrote:
>
>
> Hi,
>     I am looking for a neural software framework for
trying/modelling
> out my neural theories and learning algorithms and test them out.
I
> want to be able to spend more time just modelling/testing my
theories
> then having to develop the whole software framework myself. Any
suggestions?
> I know C/C++/Java. Something based on that would be great. Don't
want
> to have to learn something like MathLab groundup.
>
>     I specifically am looking for feed back from someone who has
worked
> with NeuroSolutions Developors Level software package from Neuro
> Dimensions Inc.
> Specifically my interest  in the package is in its capability to
model
>  new neuron theories and learning algorithms  that I may want to
play
> with. The software says it allows new neural components and
learning
> algorithms to be implemented.
>
> I just want a feel from people who have used it, regarding its
> flexibility and more importantly what its limitations are. Will
help me
> make a educated decision to picking a software framework.
>
> Thx,
> Sarvi
>
>
> [Non-text portions of this message have been removed]

Hi all,

I am a senior recruiter with Hollister in Boston.  I have been
recruiting here for over 8 years, and am working on two new searches
for one of my pharmaceutical/biotech clients that might be of
interest to some of you or your colleagues.

Both are software analysts, one focusing on a background with Machine
Learning, and the other with SMILES and cheminformatics.

Descriptions for both are below.  Please call with any questions and
send resumes directly to me at pclaughton@...

Thanks, and have a great day!

Sr./Lead Scientific Software Analyst-Machine Learning
Salary-open
Location-Cambridge

Required Skills:

-Machine Learning training and experience
-chemisty background

Education requirements:

-MS in CS

Job description:

The Senior/Lead Scientific Software Analyst will develop novel tools
and techniques for the searching and analysis
of chemical intellectual property.

Responsibilities:

-Work with modelers, software developers, and the legal group to
create the next generation of intellectual property
tools.
-research complex searching algorithims for chemical intellectual
property
-create new tools for the practical curation of generic structures
-develop novel visualization techniques that allow researchers and
legal staff to understand chemical IP coverage.

Sr./Lead Scientific Software Analyst-SMILES expertise
Salary-open
Location-Cambridge

Required Skills:

-Extensive experience with SMILES chemical representation language
-Basic ablility to program in any language
-Strong cheminformatics background

Education requirements:

-MS in CS or Chemistry preferred

Job description:

The Senior/Lead Scientific Software Analyst will develop novel tools
and techniques for the searching and analysis
of chemical intellectual property.

Responsibilities:

-Work with modelers, software developers, and the legal group to
create the next generation of intellectual property
tools.
-research complex searching algorithims for chemical intellectual
property
-create new tools for the practical curation of generic structures
-develop novel visualization techniques that allow researchers and
legal staff to understand chemical IP coverage.

Pam Claughton
Hollister
31 Milk ST.
Boston, MA 02109
617-654-0237
pclaughton@...

Hi everyone I've been a member of this groups for
quite sometime now and have been regularly reading all
the interesting ideas and topics that are discussed
however this is the first time I've written to it.
   Anyway I'm an undergrad student studying Information
systems and I'm currently working on a project which
attempts to implements Neural network architectures in
a distributed computing envoirment....and would really
appreciate it if anyone could just guide me as to how
i should go about implementing this idea in a
practical envoirment as i have still to figure out
where such a system could be applied ...I would also
appreciate it if someone could tell me how it is one
gets papers published(i.e what is the exact
procedure)...and of course any new insights on this
idea are always welcome ..

thanking everyone
Suranjit

__________________________________________________
Do you Yahoo!?
Yahoo! Tax Center - forms, calculators, tips, more
http://taxes.yahoo.com/

Dear member of this list,

Science & Consciousness Review ( http://psych.pomona.edu/scr ) has released
new articles and reviews:
_______________________

EDITORIAL REVIEW --- FOURTH COMMENTARY ON CRICK & KOCH

A computer-based model of Crick and Koch’s Framework for Consciousness
By Stan Franklin

"Crick and Koch explain that a framework must not be confused with a set of
hypotheses. Rather a framework for consciousness offers a point of view from
which to address the problems of consciousness. It’s intended to guide
research. A good framework should fit within current scientific knowledge
reasonably well and should be roughly correct." In this fourth commentary on
the Crick & Koch framework article, Stan Franklin implements their views in
Artificial Intelligence.

Full text at: http://psych.pomona.edu/scr/editorials/20030301.html

*** In February and March 2003, Science & Consciousness Review publishes
editorial commentaries
on Crick & Koch's recent article in Nature Neuroscience.
_______________________

LATEST HEADLINES

- Memory for rejected distractors in visual search?
- Using inattentional blindness as an operational definition of unattended:
The case of surface completion
- BOOK: Attention and Implicit Learning
- BOOK REVIEWS From Metapsychology
- Love arises in the unconscious

See NEWS IN BRIEF at: http://psych.pomona.edu/scr/more_news.html
_______________________

PREVIOUS ARTICLES

Consciousness lost: The lightning storm of seizures
at: http://psych.pomona.edu/scr/news/articles/20030102.html

Changing or escaping the self
at: http://psych.pomona.edu/scr/news/articles/20030101.html

Are we ever unconscious?
at: http://psych.pomona.edu/scr/news/articles/20021201.html

Visit our archives at:
http://psych.pomona.edu/scr/archive.html
_______________________

SEND YOUR CONTRIBUTION

For students, teachers, scientists, and all other fans of consciousness...
Send your news summaries to us.

See more at: http://psych.pomona.edu/scr/author_instructions.html


Sincerely,

Thomas Zoëga Ramsøy
Managing Editor

hello,
i need help regarding the optical flow calculations in vision related
problems.
suppose i am given 10 images (Yosemite) and i am required to find out
the optical flow ... how do i do it?

i understand that, given two images, the optical flow can be either
calculated by maximising the cross-correlation (or minimising the sum-
of-squared differences) or by using the fundas of departure from
smoothness, regularisation constant etc.

now what i am confused about is ... will using just two images of the
given ten give a comfortable estimation of optical flow. or else if i
wish to use all 10 images, how do i do it?

please reply soon ...
thanx
srivani

In simple optical flow terms, you calculate the optical flow between
2 images.  therefore if you have 10 and use them all, you will get
10 different optical flow frames.  It is up to you which you use.
If the movement is constant, you might try the first and last, but
there are other issues with optical flow implementation such as
window size, maximum displacement.

An excellent source for optical flow info or image processing in
general is 'Feature extraction and image processing' by M.Nixon.



--- In artificialintelligencegroup@yahoogroups.com, "srivaninarra"
<srivaninarra@y...> wrote:
> hello,
> i need help regarding the optical flow calculations in vision
related
> problems.
> suppose i am given 10 images (Yosemite) and i am required to find
out
> the optical flow ... how do i do it?
>
> i understand that, given two images, the optical flow can be
either
> calculated by maximising the cross-correlation (or minimising the
sum-
> of-squared differences) or by using the fundas of departure from
> smoothness, regularisation constant etc.
>
> now what i am confused about is ... will using just two images of
the
> given ten give a comfortable estimation of optical flow. or else
if i
> wish to use all 10 images, how do i do it?
>
> please reply soon ...
> thanx
> srivani

Can you expand? what sort of neural net? for what purpose in
particular?.  I imagine the motive is due to the fact that a lot of
neural net techniques take a significant amount of time to train,
but this is not particularly relevant in all applications.

A bit more information may allow us to help you.

Scott


--- In artificialintelligencegroup@yahoogroups.com, suranjit
adhikari <suranjit007@y...> wrote:
> Hi everyone I've been a member of this groups for
> quite sometime now and have been regularly reading all
> the interesting ideas and topics that are discussed
> however this is the first time I've written to it.
>   Anyway I'm an undergrad student studying Information
> systems and I'm currently working on a project which
> attempts to implements Neural network architectures in
> a distributed computing envoirment....and would really
> appreciate it if anyone could just guide me as to how
> i should go about implementing this idea in a
> practical envoirment as i have still to figure out
> where such a system could be applied ...I would also
> appreciate it if someone could tell me how it is one
> gets papers published(i.e what is the exact
> procedure)...and of course any new insights on this
> idea are always welcome ..
>
> thanking everyone
> Suranjit
>
> __________________________________________________
> Do you Yahoo!?
> Yahoo! Tax Center - forms, calculators, tips, more
> http://taxes.yahoo.com/

Well i'm glad that someone decided to reply ...What
i'm basically trying to do is build a Neural network
architecture which will convert the network o computer
to acts as neural net allowing the combined processing
power of the linked computers to solve complex
problems ..actually its still very early days ..and
right now i'm just trying to figure out wheather
implementing the neural net with different layers
present on different computers is even worthwhile
enough to pursue and thats where i need some
assistance ...i need to know wheather there are any
existing areas where such a system could be
used....also i'm trying to build a small software that
when downloaded will allow the computer to act as one
node of a perceptron allwing it to be interconnected
with others who have downloaded the software and
allowing it to be trained to solve different task
relevant problems.........
      So basically i'm just looking for ideas where
something like this would be useful....

thanking everybody
Suranjit

__________________________________________________
Do you Yahoo!?
Yahoo! Tax Center - forms, calculators, tips, more
http://taxes.yahoo.com/

It looks to be good idea. Much work is/has been done
on such softwares but in different prespective.
e.g. there are s/w related to some Space-research that
provide such property.
However if we can over come the communication speed
bariers of different computers Your IDEA is
achieveable + worth while.

--- suranjit adhikari <suranjit007@...> wrote:
> Well i'm glad that someone decided to reply ...What
> i'm basically trying to do is build a Neural network
> architecture which will convert the network o
> computer
> to acts as neural net allowing the combined
> processing
> power of the linked computers to solve complex
> problems ..actually its still very early days ..and
> right now i'm just trying to figure out wheather
> implementing the neural net with different layers
> present on different computers is even worthwhile
> enough to pursue and thats where i need some
> assistance ...i need to know wheather there are any
> existing areas where such a system could be
> used....also i'm trying to build a small software
> that
> when downloaded will allow the computer to act as
> one
> node of a perceptron allwing it to be interconnected
> with others who have downloaded the software and
> allowing it to be trained to solve different task
> relevant problems.........
>      So basically i'm just looking for ideas where
> something like this would be useful....
>
> thanking everybody
> Suranjit
>
> __________________________________________________
> Do you Yahoo!?
> Yahoo! Tax Center - forms, calculators, tips, more
> http://taxes.yahoo.com/
>


__________________________________________________
Do you Yahoo!?
Yahoo! Tax Center - forms, calculators, tips, more
http://taxes.yahoo.com/

--- In artificialintelligencegroup@yahoogroups.com, "digitaldel2000"
<scott.page@s...> wrote:
"I imagine the motive is due to the fact that a lot of neural net
techniques take a significant amount of time to train, ..."

What do you consider "a significant amount of time"?  I have trained
neural networks on fairly large data sets (hundreds of thousands of
examples or more with about a hundred independent variables) can be
trained in minutes on typical PC hardware.  In my experience, data
preparation and analysis take far more time than actual model
building, and this is echoed by other people in data mining.

--- In artificialintelligencegroup@yahoogroups.com, predictorx
<no_reply@y...> wrote:

>  I have trained
> neural networks on fairly large data sets (hundreds of thousands of
> examples or more with about a hundred independent variables) can be
> trained in minutes on typical PC hardware.  In my experience, data
> preparation and analysis take far more time than actual model
> building, and this is echoed by other people in data mining.

I am not a ANN enthusiast, so could you tell me what kind of results
your thousands of examples have produced.  My criticism of ANN is
that they are not able to detect precursors of significant events if
those precursors are sufficiently hidden within the complexities of
the events.

Jim Bromer

----- Original Message -----
   From: jim_bromer
   To: artificialintelligencegroup@yahoogroups.com
   Sent: Saturday, March 08, 2003 4:51 AM
   Subject: [Artificial Intelligence Group] Re: Need Assistance in Neural
Networks


   --- In artificialintelligencegroup@yahoogroups.com, predictorx
   <no_reply@y...> wrote:

   >  I have trained
   > neural networks on fairly large data sets (hundreds of thousands of
   > examples or more with about a hundred independent variables) can be
   > trained in minutes on typical PC hardware.  In my experience, data
   > preparation and analysis take far more time than actual model
   > building, and this is echoed by other people in data mining.

   -I am not a ANN enthusiast, so could you tell me what kind of results
   -your thousands of examples have produced.  My criticism of ANN is
   -that they are not able to detect precursors of significant events if
   -those precursors are sufficiently hidden within the complexities of
   -the events.
   ONe way to get around that problem is to give the net a bit of a short-term
memory. Actually, you can begin to change the input parameters to be longer bits
of information, and hopefully then your net will recognize the precursor to
significant events because the net sees it as one entire input.  However, if the
precursor is sufficiently hidden within the data, you might be asking too much
of the net to be able to recognize the precursor.

   Troy


[Non-text portions of this message have been removed]

Well, it is all dependant on the machine you use, but Ive sometimes
had to go downstairs, watch some tv, have a cuppa, only to come back
a few hours later and find its still not finished

But then again some techniques are particularly slow, especially
when you are doing stuff like L-O-O with a non-linear technnique

But yes, you are right, the data preperation takes the most time,
along with writing the code (if you haven't nicked it from someone
else).



--- In artificialintelligencegroup@yahoogroups.com, predictorx
<no_reply@y...> wrote:
> --- In
artificialintelligencegroup@yahoogroups.com, "digitaldel2000"
> <scott.page@s...> wrote:
> "I imagine the motive is due to the fact that a lot of neural net
> techniques take a significant amount of time to train, ..."
>
> What do you consider "a significant amount of time"?  I have
trained
> neural networks on fairly large data sets (hundreds of thousands
of
> examples or more with about a hundred independent variables) can
be
> trained in minutes on typical PC hardware.  In my experience, data
> preparation and analysis take far more time than actual model
> building, and this is echoed by other people in data mining.

> I am not a ANN enthusiast, so could you tell me what kind of
results
> your thousands of examples have produced.  My criticism of ANN is
> that they are not able to detect precursors of significant events
if
> those precursors are sufficiently hidden within the complexities
of
> the events.

There is a trade-off when using NN techniques -
underfitting/overfitting.  If your model is too simple you will not
capture the true relationships within the data.  If your model is
complex you will start to model the noise on top of the desired
functions, both lead to poor generalisation.

You can achieve excellent results using NN for classification and
regression, but beware people qouting % error rates, they need to be
backed up with statistical analysis (ie what is the probability we
could get the same results by random guessin? ... confidence
intervals and such like are also usefull) It is also heavily
dependant on the data set in question....



>
> Jim Bromer

We've designed a new walking robot you might be interested in. It
comes in three forms. The first kit transforms a BOE-BOT from
Parallax into a walker. we also have a pure mechanical kit ( No
Brains just the Brawn) and a full robot kit. This robot use a crank
drive so it is very easy to program. It runs standard BOE-BOT code.
Check it out at www.easybot.net

Thanks
Mike B
Easybot guy

Where can I find downloadable
free AI books or textbooks in pdf format.


__________________________________________________
Do you Yahoo!?
Yahoo! Web Hosting - establish your business online
http://webhosting.yahoo.com

you could start with the NIPS page. Not sure how much is free, but you
can find some things.

http://nips.djvuzone.org/frontpage.html

Hi guys im a beginner in the field of neural networks. Currentley im planning to
build a network for charachter recognition using back propagation. I know this
is not the best network to choose but i still im a beginner. Well my question is


How to calculate in advance the no. of hidden layer to be used and the no of
indidivisual hidden layer units to be used for the efficient training of the
net?


ive tried but i clould not find a farmula. Is there a farmula for that ?


Ayaz



---------------------------------
Do you Yahoo!?
Yahoo! Web Hosting - establish your business online

[Non-text portions of this message have been removed]

Hi,

One thing to consider is this - your image needs to be detailed
enough so that you can tell the difference between two subjects.  I
assume you are thinking of using a neural network to classify the
images? or have you another idea?  If yes, then look up face
recongition (eigenfaces and feature comparison are two popular
methods).  This will give you some idea of how that bit works.  I
imagine the 'heat map' image or whatever you like to call it will be
particularly hard to classify although maybe I've seen predator too
many times....

Scott


--- In artificialintelligencegroup@yahoogroups.com, Troy
<tdkelley1@c...> wrote:
> I am guessing that not a lot of people on this list know very much
about
> heat signatures since that is not a typical subject matter for
> artificial intelligence.  If you want to detect heat the typical
method
> is to use an infrared sensor.  If you search the net for infrared
> sensors I am sure you will find plenty of information about that
> technology.
>
> Troy
>
> Anbarasan wrote:
>
> >  Respective Group Members,
> >
> >                                              This is my second
mail,
> > firstly
> > I've message is about the topic I've selected as my paper(heat
pattern
> >
> > reorganization). Now I need your suggestions still more
specifically.
> >
> >
> >
> >                                               First of all I
coined a
> > word
> > radiation pattern sensing in humans, and at present technology
detects
> > heat
> > than other patterns in a better way. And I have some proof that
no
> > humans
> > radiate same heat pattern and some amount of pattern will be
radiating
> > from
> > each part of the body and it is not the amount of heat radiated.
Of
> > course
> > the amount of heat radiated will differ by even taking a cup of
tea,
> > but the
> > pattern wont according to me and my aim is to sense the pattern
only
> > not the
> > amount but much peoples not agreeing with this, I need your
suggestion
> >
> > please.
> >
> >
> >
> >                                             And if this is
possible I
> > need
> > idea that what sort of sensor can be used to detect, and the
distance
> > of the
> > subject? And if there is any special thing which have to kept in
mind
> > in
> > this process? As I said in my first message this is not my thesis
> > project I
> > have no one to guide so please help me with information. Do
reply.
> >
> >
> >
> >
> >
> > Yours Trustable,
> >
> > Anbarasan.
> >
> >
> > -----------
> > If This Mail Initiate You To Know Much About Me Just Visit,
> > www.arunalab.com/anbu/
> >
> >
> >
> >                    Yahoo! Groups Sponsor
>                         ADVERTISEMENT
>
>
> >
> > To unsubscribe from this group, send an email to:
> > artificialintelligencegroup-unsubscribe@yahoogroups.com
> >
> >
> >
> > Your use of Yahoo! Groups is subject to the Yahoo! Terms of
Service.
>
>
> [Non-text portions of this message have been removed]

hello AIG :

I'm new member in your group, and I want to greet you,

I study now Expert Systems, if you know any website or e-book  which can support
me please advice .

best regards,
Marwa


[Non-text portions of this message have been removed]

Hi Marwa and all,
Look at the following seciton about Expert System resources at AAAI:
http://www.aaai.org/AITopics/html/expert.html    You can find there tutorials
and introductory texts.
Moreover, If you need a software to practice then CLIPS is one of the most
popular and strongest Expert system shells. You can find all the details
including: package, documentaiton, user guide, etc. at : 
http://www.ghg.net/clips/CLIPS.html

Good luck.
Amr.
  Daghstani <mdelec@...> wrote:hello AIG :

I'm new member in your group, and I want to greet you,

I study now Expert Systems, if you know any website or e-book which can support
me please advice .

best regards,
Marwa


[Non-text portions of this message have been removed]


To unsubscribe from this group, send an email to:
artificialintelligencegroup-unsubscribe@yahoogroups.com



Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/




---------------------------------
Do you Yahoo!?
Yahoo! Web Hosting - establish your business online

[Non-text portions of this message have been removed]

Hi all

i am doing a project on character recognition and i am using back-
prapogation algorithm for training the net. sample data set is
consist of bmp pics of characters. each character size is 5x7 pixels
so the input layer contains 35 units. i first trained two characters
so the output layer has 2 units and i have made only one hidden layer
which has 20 units. this network works well and trained with noise
data. but things become worst when i increse the output units with 26
for all alphabets and hidden layer to 30 units. this new network is
not trainning with all other parameters are same. i waited 2 to 3
days for its training but nothing is happening.

my question is that how long it takes to train the network of this
kind(input units=35, hidden units = 30 , output units =26)?

wiating for reply

Touseef Liaqat

present there companies in london...listed and non listed..china counts

for more than 60 percent of the investments in china...in new technologies..

for more go to.....http://www.hkc22.de/html/chinainvestor.html

thank you...or ask ..for more...

h.k..



---------------------------------
Gesendet von http://mail.yahoo.de.
Dynamisch und innovativ - der neuen Renault Mégane.
Hier geht es zu Ihrer kostenlosen Probefahrt!.

[Non-text portions of this message have been removed]

We are in the process of developing a true general artificial intelligence.
We are 2 years into this privately funded project and have made significant
progress. We are seeking additional programmers to assist with the
development. Programming language is C#. Compensation is primarily in the
form of stock with salaries limited to subsistence levels. It is a
challenging and interesting project that is lots of fun to work on and stands
to make a big difference in the world. For more information on the project
please go to our website.

<A HREF="http://www.adaptiveai.com/">http://www.adaptiveai.com/</A>

and contact Peter Voss at:

peter@... <peter@...>



[Non-text portions of this message have been removed]

****  HIS'03 - First Call for Papers ****

3rd International Conference on Hybrid Intelligent Systems (HIS'03)

December 14 - 17, 2003
Melbourne, Australia
URL: http://his03.hybridsystem.com

********************************************************************
Deadline for Paper Submission: July 15, 2003
********************************************************************

HIS'02 is technically co-sponsored by:
- IEEE Systems, Man and Cybernetics
- The World Federation on Soft Computing
- International Fuzzy Systems Association
- Australian Computer Society
- European Neural Network Society
- European Society for Fuzzy Logic and Technology
- European Network of Excellence in Evolutionary Computing
- IOS Press

Hybridization of intelligent systems is a promising research field of
modern
artificial/computational intelligence concerned with the development
of the next generation of intelligent systems. A fundamental stimulus
to the investigations of Hybrid Intelligent Systems (HIS) is the
awareness in the academic communities that combined approaches will
be necessary if the remaining tough problems in
artificial/computational intelligence are to be solved. Recently,
hybrid intelligent systems are getting popular due to their
capabilities in handling several real world complexities involving
imprecision, uncertainty and vagueness. HIS'03 builds on the success
of last year's. HIS'02 was held in Santiago de Chile, 01-04 December
2002 and attracted participants from over 35 countries.

HIS'03 is the third International conference that brings together
researchers, developers, practitioners, and users of soft computing,
computational intelligence, agents, logic programming, and several
other intelligent computing techniques. The aim of HIS'03 is to serve
as a forum to present current and future work as well as to exchange
research ideas in this field.

HIS'03 invites authors to submit their original and unpublished work
that demonstrate current research using soft computing/ computational
intelligence and other intelligent computing techniques and their
applications in science, technology, business and commercial.

HIS'03 will focus on the following themes:

Hybrid Intelligent Systems Architectures and Applications
Soft Computing for Image and Signal Processing
Intelligent Internet Modeling, Communication and networking
Intelligent Data mining
Intelligent Business Systems
Soft Computing for Control and Automation
Intelligent agents
Knowledge management
Special topics

We invite you to submit a:

- full paper of 8 to 10 pages (Letter or A4 paper) for oral
presentation
- proposal to organize a technical session and/or workshop (see the
Call
for Events Proposals in the conference Web page for more information).

Submitted papers have to be original, containing new and original
results.
The proceedings of the Conference will be published by IOS Press, The
Netherlands. It is assumed that all accepted manuscripts would be
presented at the conference. All accepted papers must be accompanied
by a full paid registration to appear in the proceedings. All full
papers are to be submitted in PDF electronically via the web site.
Hard copies should be sent only if electronic submission is not
possible. All papers will be peer reviewed by at least two
independent referees of the international program committee of
HIS'03.


Topics of interest include but not limited to:

Hybrid Intelligent Systems Architectures and Applications

Interactions between neural networks and fuzzy inference systems
Artificial neural networks and learning techniques
(supervised /unsupervised / reinforcement learning)
Artificial neural network optimization using global optimization
techniques
Fuzzy clustering algorithms and optimization techniques
Fuzzy inference system optimization using global optimization
algorithms
Evolutionary computation (genetic algorithms, genetic programming,
evolution strategies, grammatical evolution etc.)
Support vector machines, rough sets, Bayesian networks, probabilistic
reasoning, minimum message length, etc.
Hybrid computing using neural networks - fuzzy systems - evolutionary
algorithms
Hybrid optimization techniques (evolutionary algorithms, simulated
annealing, tabu search, GRASP etc.)
Hybrid of soft computing and statistical learning techniques
Intelligent agents (architectures, environments, adaptation/ learning
and knowledge management)
Models using inductive logic programming, logic synthesis,
grammatical inference, case-based reasoning etc.
Autonomic computing

Soft Computing for Image and Signal Processing

Image generation, acquisition, and processing
Virtual reality
Scene and object modeling
Document image understanding
Image algebra
Design and implementation of intelligent signal processing systems
Image, video and multidimensional signal processing
Speech processing
Features and classification
Texture analysis
Stereoscopic vision
Shape processing
Object recognition
Image and video retrieval
Image and video compression

Intelligent Internet Modeling, Communication and networking

Web intelligence
Intelligent Search engines
Database querying
XML mining
Intelligent networking between Web Sites
Network security, intrusion detection
Intelligent agents and interfaces for personalization and adaptivity
Intelligent tutoring systems on the WWW
Adaptive hypermedia systems
Agents for digital cities, virtual communities and agent societies
Web Services modeling
UDDI and SOAP enhancements
Network Management
Mobile Internet
Wireless multimedia, QoS adaptation
WAP, mobile e-commerce
Distributed algorithms for wireless networks

Intelligent Data mining

Discovering patterns in continuous data
Uncertainty management for data mining
Clustering algorithms and applications
Classification trees
Mining time series
Mining in a Mobile Environment
Statistical Considerations in Learning
XML Mining
Text Mining
Distributed Data Mining

Intelligent Business Systems

e-learning, e-commerce, e-business, e-finance
Risk management
Derivatives pricing
Portfolio management and asset allocation
Stock market, forex market analysis, dynamics and simulation
Hedging, trading & arbitrage strategies
Financial modeling
Computational economics
Intelligent management
Multicriteria decision making

Soft Computing for Control and Automation

Mathematical modeling and analysis of complex systems
Soft computing/computational intelligence in control systems
Knowledge based control systems
Adaptive control systems
Control applications in robotics manufacturing, process control,
industrial systems, automotive, vehicular systems, spacecraft and so
on

Intelligent Agents

Adaptation and learning
Agent architectures and communication languages
Conversational agents
Coordinating multiple agents
Designing agent systems - methodologies & software engineering
Evolution of agents
Knowledge acquisition and management
Models of emotion, motivation, or personality
Multi-agent communication, coordination, and collaboration

Knowledge Management

Application of knowledge representation techniques to semantic
modeling
Development and management of heterogeneous knowledge bases
Automatic acquisition of data and knowledge bases (especially raw
text)
Performance evaluation
Data and knowledge sharing
Cooperation in heterogeneous systems
Domain modeling and ontology-building
Concurrent engineering and computer integrated manufacturing
Digital Libraries
Multimedia Databases.

Affective Computing

Kansei Information Processing
Models, theories and taxonomies of emotions
Emotion synthesis
Emergent emotion and emotional behavior
Emotion recognition
Emotional expression
Learning and emotions
Emotions and perception
Emotions and social interaction
Emotional Agent Architecture
Embodiment of emotional systems
Emotions and other affective phenomena (moods, temperament, and
personality)
Applications: Synthetic characters, robots, computational theater,
interface assistants, pedagogical agents, music, art, education,
entertainment, multi-agent systems, bio-feedback, etc.


Important Dates

Deadline for workshop/ sessions/ tutorial proposals: 02 June'03
Deadline for paper submission (full paper): 15 July'03
Notification of acceptance : 25 August'03
Deadline for camera ready papers and authors' registration: 10
September'03
HIS'03 conference in Melbourne: 14-17, December'03

********************************************************
HIS03 Organization

Honorary Chair
David B. Fogel, Natural Selection Inc, USA

General Chairs
Ajith Abraham, Oklahoma State University, USA
Mario Köppen, Fraunhofer IPK Berlin, Germany

Program Chairs
Kate Smith, Monash University, Australia
Lakhmi Jain, University of South Australia, Australia

Advisory Board
Javier Ruiz-del-Solar, Universidad de Chile, Chile
Fabio Abbattista , Universita di Bari, Italy
Janusz Kacprzyk, Polish Academy of Sciences, Poland
Antony Satyadas, IBM Corporation, Cambridge, USA

Local Organizing Committee
Damminda Alahakoon, Monash University, Australia
Vincent Lee, Monash University, Australia
Ai Cheo Yeo, Monash University, Australia

Web Chair
Fabio Zambetta, Universita di Bari, Italy

Special Sessions Chair
Xiao-Zhi Gao , Helsinki University of Technology, Finland

Finance Coordinator
Diane West, Monash University, Australia

International Technical Committee
Janos Abonyi, University of Veszprem, Hungary
Jose Ramon Alvarez Sanchez, UNED, Spain
Bruno Apolloni, Università degli Studi di Milano, Italy
Akira Asano, Hiroshima University, Japan
Vijayan Asari, Old Dominion University, USA
Bernard de Baets, Ghent University, Belgium
Ricardo Baeza-Yates, Universidad de Chile, Santiago, Chile
Sung Bae Cho, Yonsei University, Korea
Vic Ciesielski, RMIT University, Australia
Carlos A. Coello Coello, CINVESTAV-IPN, Mexico
David W. Corne, University of Reading, UK
Paulo Jose da Costa Branco, Instituto Superior Técnico, Lisboa,
Portugal
Evgenia Dimitriadou, Vienna University of Technology, Austria
Yasuhiko Dote, Muroran Institute of Technology, Japan
Mark Embrechts, Rensselaer Polytechnic Institute, USA
Anna Maria Fanelli, Universita' degli Studi di Bari, Italy
Luciano da Fontoura Costa, University of Sao Paulo, Brazil
Katrin Franke, Fraunhofer IPK Berlin, Germany
Matjaz Gams, Jozef Stefan Institute, Ljubljana, Slovenia
Xiao-Zhi Gao, Helsinki University of Technology, Finland
Tom Gedeon, Murdoch University, Australia
Jerzy W. Grzymala-Busse, University of Kansas, USA
Saman K. Halgamuge, The University of Melbourne, Australia
Tim Hendtlass, Swinburne University of Technology, Australia
Francisco Herrera, University of Granada, Spain
Frank Hoffmann, University Dortmund, Germany
Ravi Jain, James Cook University, Australia
Robert John, De Montfort University, UK
Janusz Kacprzyk, Systems Research Institute, Warsaw, Poland
Etienne Kerre, Ghent University, Belgium
Frank Klawonn, University of Applied Sciences
Braunschweig/Wolfenbuettel, Germany
Andreas Koenig, University of Kaiserslautern, Germany
Amit Konar, Jadavpur University, India
William B. Langdon, University College, London, UK
Andre Ponce de Leon Ferreira de Carvalho, University of Sao Paulo,
Brazil
Sebastian Lozano, Escuela Superior de Ingenieros, Sevilla, Spain
Luis Magdalena, Universidad Politecnica de Madrid, Spain
Maria do Carmo Nicoletti, Universidade Federal de São Carlos, Brazil
Erkki Oja, Helsinki University of Technology, Finland
Antônio de Pádua Braga, Federal University of Minas Gerais, Brazil
Sankar K. Pal, Indian Statistical Institute, India
Nikhil R. Pal, Indian Statistical Institute, India
Marcin Paprzycki, Oklahoma State University, USA
Witold Pedrycz, University of Alberta, Canada
Guenther Raidl, Vienna University of Technology, Austria
Rajkumar Roy, Cranfield University, UK
Javier Ruiz-del-Solar, Universidad de Chile, Santiago, Chile
Antony Satyadas, IBM Corporation, Cambridge, USA
Giovanni Semeraro, Universita' degli Studi di Bari, Italy
Dharmendhra Sharma, University of Canberra, Australia
Clarence W. de Silva, University of British Columbia, Canada
Aureli Soria Frisch, Fraunhofer IPK Berlin, Germany
Branko Soucek, Iris, Bari, Italy
Martin Stytz, Air Force Research Laboratory, USA
Andrew Sung, New Mexico Tech, USA
Jarno Tanskanen, University of Kuopio, Finland
Olgierd Unold, Wroclaw University of Technology, Poland
Marley Maria B.R. Vellasco, PUC-Rio, Brasilia
Lipo Wang, Nanyang Technological University, Singapore
Geoffrey Webb, Monash University, Australia
Richard Weber, Universidad de Chile, Santiago, Chile
Kaori Yoshida, Kyushu Institute of Technology, Japan
Chengqi Zhang, University of Technology, Sydney, Australia
Zili Zhang, Deakin University, Australia

"jim_bromer" wrote:
"I am not a ANN enthusiast, so could you tell me what kind of results
your thousands of examples have produced."

Yes.  As an example, I used to produce predictive models for mutual
fund companies using neural networks.  Clients would provide customer
(or sometimes broker) data which ranged from 30,000 to 1.5 million
records (typical range: 250,000 - 700,00 records: one record per
customer, each being a single example for modeling), with a few
hundred attributes (demographic data, monthly transaction summaries,
etc.).  After screening out obviously useless or redundant variables
(such as ones that didn't vary), the pre-processed data set had nearly
as many records as the original data set (some customers were filtered
out because they didn't have enough money to bother analyzing) and
somewhere between 75 and 250 variables (typically something close to
100).

Feedforward neural network models were constructed on ordinary (at
that time) desktop hardware (about 800MHz PC, with extra RAM: 512
megabytes) using the usual variations on backpropagation (momentum
term, etc.).  The neural networks generally included all input
variables and had between 10 and 150 hidden neurons (15-30 being
typical).  Models were trained in less than 20 minutes for data sets
with about 250,000 cases.

These models predicted customer defection (closed accounts) three
months in advance.  Performance (when class accuracies were balanced)
was usually in the 70%-85% range, with low 80s being fairly common- a
smashing success for these clients.

I'll note that the biggest difference I've observed between my own
work using neural networks and things reported in the literature and
online is the number of training passes needed to train a neural
network.  I found that networks frequently reached optimal training
(test error bottomed out) in 10 passes or less (keeping in mind that a
single pass over several hundred thousand cases is still a lot of
training).




"My criticism of ANN is that they are not able to detect precursors of
significant events if those precursors are sufficiently hidden within
the complexities of the events."

Of course that is true, by definition.  The first part of the above
sentence, "not able to detect precursors of significant events", is
merely a restatement of the second part, "precursors are sufficiently
hidden within the complexities of the events", and would be true of
any system.  To my mind the real issue is how often this happens in
practice.  In my experience, neural networks (and other empirical
modeling systems) often (though not always) provide a desireable
solution to a data anlysis problem.

Why are you using 26 units?  Why not use 5 units to represent a binary number
between 0 and 31 instead of 26 for 0 to 25? (A=00000, B=00001, C=00010, ...,
Z=11010)
  Touseef Liaqat <paramount01us@...> wrote:Hi all

i am doing a project on character recognition and i am using back-
prapogation algorithm for training the net. sample data set is
consist of bmp pics of characters. each character size is 5x7 pixels
so the input layer contains 35 units. i first trained two characters
so the output layer has 2 units and i have made only one hidden layer
which has 20 units. this network works well and trained with noise
data. but things become worst when i increse the output units with 26
for all alphabets and hidden layer to 30 units. this new network is
not trainning with all other parameters are same. i waited 2 to 3
days for its training but nothing is happening.

my question is that how long it takes to train the network of this
kind(input units=35, hidden units = 30 , output units =26)?

wiating for reply

Touseef Liaqat



Yahoo! Groups SponsorADVERTISEMENT

To unsubscribe from this group, send an email to:
artificialintelligencegroup-unsubscribe@yahoogroups.com



Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.



---------------------------------
Post your free ad now! Yahoo! Canada Personals


[Non-text portions of this message have been removed]

ayaz shahid wrote:
"How to calculate in advance the no. of hidden layer to be used and
the no of indidivisual hidden layer units to be used for the efficient
training of the net?

ive tried but i clould not find a farmula. Is there a farmula for
that?"


No, there is no such formula because it all depends on the nature of
the relationship between inputs and outputs and the quality of the
data.  Many rules of thumb have circulated online, but they are
generally worthless except perhaps as a starting point.

A problem with a single input variable and a single output variable
may involve a mapping so complicated that many hidden nodes are
required.  A problem with many inputs and many outputs may involve a
mapping so simple that only a single hidden node is needed.
Experimentation is necessary to decide what is useful in any
particular problem.

In my experience, the good news is that one need not try many
different sizes of hidden layers since performance in practice tends
to vary slowly across sizes.

"digitaldel2000" <scott.page@s...> wrote:
"Well, it is all dependant on the machine you use, but Ive sometimes
had to go downstairs, watch some tv, have a cuppa, only to come back a
few hours later and find its still not finished

But then again some techniques are particularly slow, especially when
you are doing stuff like L-O-O with a non-linear technnique"

That's true.  I should have also mentioned that the neural networks I
was describing used a simple holdout test strategy.  With the amount
of data I had at that job, there was no need for things like
leave-one-out, etc.

Regarding using a 5 bit number to represent 26 letters of the alphabet
as neural network output representation:

Kenneth Bull wrote:
"Why are you using 26 units? Why not use 5 units to represent a binary
number between 0 and 31 instead of 26 for 0 to 25? (A=00000, B=00001,
C=00010, ..., Z=11010)"


This would require the neural network to learn a much more complicated
mapping for each individual bit- in essence learning the clasification
and learning to be a 32-to-5 encoder.  Consider your bit number 4 in
the above examples you've given.  Classes "A" and "B" have bit number
4 (counting from the left) the same (zero) and classes "C" and "Z"
have it as a one.  This is arbitrary and probably does nto reflect
structural differences in the classes.

This sort of representation is suggested frequently online but is
dreadful in practice since the model needs to learn mappings to
classes and also learn to turn off those mappings in a convoluted way.

Keep in mind that the basis functions used by most artificial neural
networks in practice are simple, monotonic transfer functions.
Learning to turn "on" one of your bits requires one or more basis
functions.  Learning to turn them "off" for this arbitrary
representation may take many more.