Greetings
from NeuroDimension!
The World Leader in Neural Network Software
This issue of the NeuroDimension newsletter
highlights a new feature of NeuroSolutions 4.0 and the upcoming neural network
course.
In this issue you’ll find:
What’s New
and News?
*
Deadline Approaching for Neural Network Course Discount
* New Structure for NeuroSolutions University Site License
* Conjugate Gradient
Learning
Customer
Spotlight
*
Stream Flow Prediction
Note: You are
receiving this newsletter because you requested to stay informed concerning new
developments at NeuroDimension. If you would like to stop receiving these
newsletters, please see the bottom of this newsletter for removal instructions.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
The deadline for an early registration discount is
approaching soon for our May neural network courses in Orlando, Florida. The
courses will take place May 7-11, 2001 at the Grosvenor Resort, located in the
Walt Disney World Resort. The deadline to qualify for a 10% discount is April
1st. This session is filling up fast, so be sure to register today!
Our course format allows both novice and advanced users to
find a suitable course. Offered courses include: "Introduction to
NeuroSolutions", "Fundamentals of Neural Networks and
NeuroSolutions", and "Advanced NeuroSolutions". In addition,
each of the courses includes updates on the new features of NeuroSolutions 4.0,
geared towards each level of user.
The courses include a copy of our interactive book, Neural
and Adaptive Systems: Fundamentals Through Simulations. We are also happy
to work with attendees who would like to use their own data in the sample
projects.
For details on this new offering, or to sign-up from the
Internet, see http://www.nd.com/course/may_2001.htm
For general ND course information, see http://www.nd.com/course
For more information and samples of the interactive book,
see http://www.nd.com/products/nsbook.htm
Many of our academic customers have found the University
Site License a cost-effective way to utilize NeuroSolutions on multiple
computers. Previously, the unlimited student installations allowed by the site
license were restricted to the Educator level of the software. We have recently
restructured our University Site License so that now the student installations
can be at the Educator, Users, Consultants or Developers Lite level. The new
Site License price list can be found at:
http://www.nd.com/univsite.htm
Existing educational customers may apply their original
purchase price towards an upgrade to one of these new site license options. An
upgrade price quote can be obtained by emailing to sales@nd.com.
Note that this site license pricing is not available to corporations.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
A new feature of NeuroSolutions 4.0 is conjugate gradient
learning. NeuroSolutions uses the Scaled Conjugate Gradient implementation of
this learning technique. Second order learning methods (like Newton’s method)
use not only the slope of the performance surface but also the curvature to
adjust the weights. As an example of the power of second order methods, it is
known that linear systems always have a quadratic performance surface. Second
order methods can reach the bottom of a quadratic performance surface in one
step. Second order methods, however, are computationally very expensive.
Conjugate gradient is an approximate second order method that is an excellent
trade-off between computational complexity and increased learning speed. In
general, a conjugate gradient training epoch in NS will take twice as long as a
standard gradient descent training epoch. The conjugate gradient method,
however, will typically train in much fewer epochs and also move to a lower
final MSE. Another significant advantage of scaled conjugate gradient learning
is that it is parameterless. No need to
set learning rates or momentum terms. It automatically determines the “best”
step size at each iteration.
Conjugate gradient in NeuroSolutions, however, must be used
in batch mode. As we discussed in our August ‘99 newsletter, there are
circumstances where batch learning does not work as well as on-line. In
particular, with data sets larger than 500-1000 samples on-line learning may
train much faster. In these cases, often times on-line learning with momentum
or some other standard learning algorithm may outperform conjugate gradient
learning.
In NeuroSolutions, the “BackStaticControl” component (red
dials) controls the error calculations and gradient descent learning. On the
“BackStaticControl” Inspector, on the “Backpropagation” page, there are three
radio buttons labeled “on-line”, “batch”, and “custom”. These selections determine
how much time is spent calculating the gradient before the weights of the
system are updated (see August ‘99 for
more detail). In addition to controlling the on-line vs. batch mode of
learning, there is a panel on the right of the “Backpropagation” page that
controls the gradient search method (it is labeled “plane” and says “gradient
search” inside the box). To switch to conjugate gradient learning (or any other
method), click the “remove” button to remove the existing backpropagation
plane, then select the new gradient search method (e.g. conjugate gradient) in
the pull-down box and click “add”. This
will set-up the breadboard to use conjugate gradient learning.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Bernard B. Hsieh, USAE R&D Center
A stream flow prediction system is developed by the
Artificial Neural Networks (ANN) for addressing the flood forecasting issues of
two different scale watersheds: the Sava River, Croatia and a segment of the
lower Mississippi River. The study investigated the prediction system with
single-point river stage, upstream-downstream river flow forecasting, and
rainfall-runoff hydrological process. The study indicated that the minimum
length of river stages required achieving about 90 percent of up to 3 days
forecasting reliability was about 3 months for the Sava River. The reasonable
downstream river flow prediction from upper stream gauges was found in the Sava
River even only half year daily values were available for model training. On the Mississippi River, with 16 years
long-term daily information, the ANN can construct a very high precision river
flow forecasting system for Memphis, TN, from two upstream inputs, near the
confluence of the Ohio River, without significant rainfall contribution in this
river segment.
This is just an abstract of the application summary. The
entire summary is available at:
http://www.nd.com/application%20summaries/appsum-predict.htm
Want to have your solutions spotlighted? We strongly
encourage our customers to send their 1-2 page application summaries to submissions@nd.com so that we may post
them on our web site at: http://www.nd.com/applicationsum.htm.
In selected newsletters, we’ll spotlight a new solution and include a link for
people to get more information.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
We appreciate your feedback! Please send us your comments or
suggestions concerning this newsletter, our web site, or part of the
NeuroDimension product line. Write to us at: feedback@nd.com
Have questions about NeuroDimension products or training
services? Send your questions to: info@nd.com
Some mail readers may display an attachment at the bottom of
this newsletter. This is typically caused when the mail reader is set to
display text only. The original, formatted version is sometimes included as an
attachment.
This issue and previous issues of this newsletter are
available on the NeuroDimension web site at: http://www.nd.com/mailinglist.htm
If you would prefer not to receive these newsletters or
subsequent product updates from NeuroDimension, please reply to this letter with
the subject heading changed to the word REMOVE.
Thank you again for your support of NeuroDimension products!