Greetings
from NeuroDimension!
The World Leader in Neural Network Software
This issue of the NeuroDimension
newsletter highlights several new features introduced in the recent release of
NeuroSolutions, along with some exciting offers from NeuroDimension!
In this issue you’ll find:
What’s New
and News?
* NeuroSolutions v4.14 Upgrade
Available
* 10% Discount on Neural Network
Course
* Designing with the
NeuralExpert
* New Kohonen Access Points
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.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
NeuroSolutions v4.14 is now
available from NeuroDimension. This release addresses several minor bugs including:
* Genetic Optimization
produces too many duplicate chromosomes
* Custom Solution Wizard
project shell for Excel runs slowly when there is a single output
And, a nice enhancement:
*
Optimization log -- Double-click on the GeneticControl component to view
the optimized parameter settings during a genetic training run.
If you already have NeuroSolutions v4.0 or v4.1 installed on
your computer you can upgrade to v4.14 by downloading and running the patch
from http://www.nd.com/support/ns_patch.htm
The
complete installation program is available for download at http://www.nd.com/download.htm
The next neural network course is scheduled for November
5-9, 2001 at the Grosvenor Resort, located in the Walt Disney World Resort in
Orlando, Florida. A 10% discount is available for early registration, 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 course, or to sign-up from the Internet,
see http://www.nd.com/course/nov_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
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Over the past few years we have discussed a number of ways
to improve the performance of your networks.
NeuroSolutions 4.0 includes a new wizard called the NeuralExpert that
embeds many of these topics into a smarter, easier to use wizard. It starts by
asking you what type of problem you have to solve (“classification”, “function
approximation”, “prediction”, or “clustering”, all of which are explained in
the help file) and then intelligently selects the appropriate network for you.
It sets reasonable defaults for all of the parameters and opens intelligently
selected probes. For instance, in a classification problem the mean squared
error is not as important as the number of correct classifications. Thus, a
confusion matrix (showing the number of correct and incorrect classifications
for each class) is opened for the classification problems.
Another important feature is the ability to modify an
existing network using the wizard, instead of having to make the changes by
hand. When the NeuralExpert builds a
network, it places three buttons at the top of the breadboard. A help button
displays a help file that describes the breadboard and all of the components. A
testing button automatically loads the testing wizard. And a modify button
reloads the NeuralExpert and loads in the current status of the breadboard into
the NeuralExpert panels. This allows you to make a single change (say from 20%
cross validation to 40% cross validation) and immediately return to the
breadboard.
This facility also helps address the network size issue that
we have stressed in previous newsletters. The last panel allows you to select a
level of “network complexity”. We suggest you start with a “low complexity”
networks, evaluate its performance, then try the “medium complexity”, evaluate
its performance, etc. The lowest
complexity network that adequately solves your problem will most likely generalize
better to unseen (out of sample) data. The NeuralExpert will make intelligent
decisions about the number of PEs and number of layers for the low, medium, and
high complexity networks.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Kohonen Self-Organizing Maps (SOMs) have been a part of
NeuroSolutions since its first release in 1995. Prior to NeuroSolutions v4.0,
pure SOMs had to be manually built and the information provided by the Kohonen
Synapse had limited use. With the introduction of the NeuralExpert in v4.0, a
SOM is built for you automatically once you select "Clustering" as
the problem type and you specify the input data. The NeuralExpert also stamps
probes on all of the new access points related to SOMs. A brief definition of
each access point is outlined below. A complete description of each can be
found within the online help of both NeuroSolutions and the NeuralExpert.
Unified Distance - This access point produces the
distance from each PE to its neighbors. Looking for large distances (light
values on an ImageViewer probe, or large black squares on a Hinton probe) shows
input cluster boundaries.
Component Plane - This access point allows you to
view only the weights from a single input (from the multi-dimensional input
vector) to all the PEs, in order to see how that input varies from cluster to
cluster. For example, if there are regions in the Self-Organizing Map (SOM)
where the weights for a particular input are very high, then we can say that
the inputs found in that cluster have a high value for that input.
Frequency - This access point provides a histogram of
win frequencies for each PE, which gives information to help determine the
clustering. Because of the neighborhood restrictions on the PEs of a SOM, often
times you will have "dead PEs" (PEs that win very few competitions)
that are in "empty" areas of the input space between clusters. By
finding these dead PEs, you can locate the borders of the clustering inside
your SOM.
Quantization Metric - This access point produces the
average quantization error, which measures the "goodness" of fit of a
clustering algorithm. It is the average distance between each input and the
winning PE. If the quantization error is large, then the winning PE is not a
good representation of the input. If it is small, then the input is very close
to the winning PE.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
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!