Greetings
from NeuroDimension!
The World Leader in Neural Network Software
This issue of the NeuroDimension
newsletter highlights more new features introduced in the recent release of
NeuroSolutions, along with an announcement concerning the upcoming neural
network course!
In this issue you’ll find:
What’s New
and News?
* Neural Network Course Savings
* NeuroSolutions Now Supports
Hardware Keys
* Iterative Prediction and
Teacher Forcing
* Natural Ventilation;
Dealing with the Unpredictable
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 is quickly approaching for a 10% early
registration discount for our next series of neural network courses. The
courses are scheduled for November 5-9, 2001 at the Grosvenor Resort, located
in the Walt Disney World Resort in Orlando, Florida. Register before October
1st to receive a 10% discount on any course!
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
Many of you may remember NeuroSolutions 3’s use of a piece
of hardware called a “key” or “dongle”. This device was required to be plugged into
the parallel port of your computer in order to use the software. This system
required people to wait for the key to be mailed to them before they could use
NeuroSolutions. However, it did allow people to easily transfer licenses
between multiple computers.
NeuroSolutions 4.0 and TradingSolutions both use an
activation code system that allows people to use the software soon after it is
purchased. This has proved to be more convenient for most customers, but
slightly more difficult for people who would like to use multiple computers.
For the added convenience of our multi-computer customers,
NeuroSolutions 4.0 has been updated to support hardware keys through either the
USB ports or the parallel ports on your computers. This will allow you to use
one set of activation codes for each installed copy of the software.
Hardware keys are available for $50. Current and prospective
customers of NeuroSolutions 4.0 who are interested in hardware key support
should call NeuroDimension or send a request for more information to info@nd.com
Hardware key support for TradingSolutions will be introduced
in the near future.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Neural networks trained with backpropagation through time
are well-suited towards the task of predicting a nonlinear time series. In
NeuroSolutions, these models are the ones that use the DynamicControl
component, which are the Recurrent and the Time-Lag Recurrent networks.
NeuroSolutions 4.0 now includes an enhancement to the DynamicControl component,
which implements an algorithm called iterative prediction.
Iterative prediction is one of two standard methods to solve
a multi-step prediction problem (predicting N-steps in advance, typically in a
time series). One method is to simply use the current and previous data to
predict N-steps ahead. In this case, the goal is to identify a direct
relationship between the inputs and a value N-steps in the future.
A simple 3-step prediction could be viewed like this:
(A0 >> A3)
Iterative prediction differs from this in that it predicts
only one sample ahead. It then uses the output of this prediction as an input
to a prediction of the next sample. This cycle is continued until a prediction
N-steps ahead is made. This approach is a form of dynamic modeling, where the
overall goal is to identify the system that created the time series.
A 3-step iterative prediction would look like this:
(A0 >> A1)... (A1 >> A2)... (A2 >> A3)
It is important to note that since the inputs to each
successive prediction are generated by the previous prediction, there must be a
one-to-one relationship between the inputs and outputs.
A 3-step iterative prediction with two inputs would look
like this:
(A0, B0 >> A1, B1)... (A1, B1 >> A2, B2)... (A2,
B2 >> A3, B3)
Because errors induced by each successive prediction make it
very difficult to train an iterative predictor, teacher forcing is
generally used to simplify the training task.
Teacher forcing feeds a portion of each input exemplar from
the file and the remaining portion from the network output. The most common
approach is to start training with a large portion of the input data coming
from the input file then gradually decreasing it such that the network produces
more of the input data. Thus, you train the network to do a simple one-step
prediction, and then once it has mastered the one step prediction you require
it to do two-step prediction, etc.
A panel in the NeuralBuilder can be used to configure a
network for iterative prediction and teacher forcing. This panel will only
appear if the following conditions are met:
1. The neural model
is either "Time-Lag Recurrent" or "General Recurrent".
2. The
"Predict" switch is set (from the Training Data panel).
3. All columns are
selected as either "Skip" or "Predict" (none as
"Input").
4. The
"Delta" is greater than 1.
For more information on this panel, please see the
NeuralBuilder help page entitled "Step 5: Simulation Control".
Iterative prediction and teacher forcing can also be set up
directly through the DynamicControl Inspector.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Richard Aynsley and Regan Potangaroa
The Australian Institute of Tropical Architecture
James Cook University
Architects often design buildings, particularly in humid
tropical climates, to take advantage of natural ventilation. From a wind rose
for the general area, architects note the typical wind speeds and prevailing
wind directions for the hotter months of the year. The architect uses this
information during design in deciding the size and location for natural
ventilation openings.
If thermal comfort enhancement is the design objective, what
is the probability it will be achieved or more importantly what is the
probability it will not be achieved? Even eliminating influences such as
metabolism and clothing, there still remain strong influences from air
movement, air temperature, humidity and radiant heat gain. Careful design to
limit indoor surface temperatures can avoid indoor radiant heat gain.
This paper suggests a statistical method for estimating the
probability of natural ventilation equaling or exceeding that required for
thermal comfort based on long term 3 hourly simultaneous data on temperature,
humidity and wind. Using simultaneous temperature, humidity and wind data
avoids the difficulties of estimating multivariate probability of variables,
which are not strictly independent. Finally, preliminary work using neural
networks is introduced.
This is just an abstract of the application summary. The
entire summary is available at:
http://www.nd.com/application%20summaries/appsum-arch.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!