Greetings from NeuroDimension!
Makers of NeuroSolutions, the Neural Network Simulation Environment.
This issue of the newsletter highlights NeuroSolutions as a teaching and research tool. In fact, we use NeuroSolutions for over 100 hands-on examples in our neural network training courses. Plus, this newsletter also includes the tips and techniques that you’ve come to expect from the professionals at NeuroDimension.
In this issue you’ll find:
What’s News?
* A Few Seats Still Available For November Course
* Get Your Frequently Asked Questions Answered
Designing Neural Networks
* Testing the Power of Nonlinear Systems
NeuroSolutions Tip Box
* Using the Educator to Train High-level Breadboards
Did You Know?
* Custom Solution Wizard Example: Providing Training Status and Allowing the End-user to Stop Training
Customer Spotlight
* Modeling the Effect of Carbon Content on Hot Strength of Steels
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.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
What’s News?
A Few Seats Still Available For November Course
The next NeuroDimension course offering will be November 8-12, 1999 in Orlando, Florida. There are still a few seats available.
We continue to receive rave reviews from our attendees, in fact everyone has said that they would recommend our courses to friends. We average a score of 4.6 of 5 when attendees are asked to rate the overall quality of the course.
Our overlapping course format allows both novice and advanced users to find a suitable course. Offered courses include: "Introduction to Neural Networks and NeuroSolutions", "Advanced Neural Networks and NeuroSolutions", and "NeuroSolutions for Developers".
For details on the November offerings, or to sign-up immediately from the internet, see http://www.nd.com/course/november.htm
For general ND course information, see http://www.nd.com/course
Get Your Frequently Asked Questions Answered
This month we would like to highlight a useful resource for technical support on our web site. An easy way to receive quick answers to routine questions about NeuroSolutions is to visit our online Frequently Asked Questions (FAQ). Our technical support team has compiled our FAQ with detailed solutions to over 40 of the most common problems our customers have experienced. We also offer the FAQ in Word format for those of you who would like to download the FAQ for use off-line.
http://www.nd.com/support/nsfaq.htm
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Designing Neural Networks
This Month: Testing the Power of Nonlinear Systems
The true power of most neural network topologies is derived from their nonlinear processing of the data. Neural networks harness the power of nonlinear "machines" while maintaining their ability to be trained. Unfortunately, it is easy to blindly apply neural networks to a solution and never determine if the neural networks are advantageous to a particular problem. It can be very useful to compare any neural network solution with an optimal linear solution to determine if the additional complexity of the nonlinear system is worthwhile.
In NeuroSolutions, it is very easy to do this for multi-layer perceptrons (MLPs) and similar networks (TLFNs, modular, etc.). There are two methods for creating a linear network based on a nonlinear network. One method is to replace all the nonlinear axons in the network (tanh and sigmoid axons) with bias axons. Switching axons is very simple, just open the Axon palette, select the bias axon, and then stamp it on the nonlinear axon. This will replace the nonlinear axon with a linear one with exactly the same properties (number of PEs, probes, etc.). Do this to each nonlinear axon to create an equivalent linear network. A second method for neural wizard breadboards would be to create a second network with the neural wizard exactly like the first one, except for each layer of the network select "bias axon" as your "PE transfer" function. This will create an identical linear network.
Train these linear networks exactly as you did the nonlinear network and compare the results. Usually you will see a marked improvement with the neural network over its linear equivalent. If you don't, a few possible reasons are:
* your problem may be simple enough to be solved with a linear network
* your neural network is not being trained correctly
* your data may not contain enough information to appropriately solve the problem
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
NeuroSolutions Tip Box
This Month: Using the Educator to Train High-level Breadboards
There are occasionally customers that would like to run simulations of sophisticated networks on multiple machines, but do not want to have to purchase a high-end license for each machine. University professors with teaching or research labs often fall into this category. The temporary license feature of NeuroSolutions offers a viable alternative.
Suppose you have purchased one copy of the Consultants level and five copies of the Educator. You have used the Consultants license to build an elaborate neural network and you would like to train the weights of this topology on the other five machines, using different parameters for each. Once you copy the breadboard to those machines, the networks will run just as if they were licensed for the Consultants level. However, there are two restrictions: 1) you cannot add or remove components, and 2) the breadboard expires after 30 days. The temporary license can be renewed by re-saving the breadboard with the Consultants license.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Did You Know?
This Month: Custom Solution Wizard Example:
Providing Training Status and Allowing the End-user to Stop Training
Many of our users have requested an example of how they can provide status and/or the ability to stop training when using a neural network DLL generated by the Custom Solution Wizard. An example that demonstrates both of these functions complete with source code is now available as a self-extracting executable from our web site at http://www.nd.com/newsletter/StopTrainingExample.exe.
This is a stand-alone example and does not require you to have the Custom Solution Wizard installed or to be a licensed user of the Custom Solution Wizard. The example program displays the current epoch number and mean-squared error while training. It also allows the user to stop the training or randomize the weights while training. If you would like to add any of this functionality to your neural network enabled application, please download the example by clicking the link above.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Customer Spotlight
This Month: Modeling the Effect of Carbon Content on Hot Strength of Steels
L.X. Kong and P.D. Hodgson
School of Engineering and Technology, Deakin University, Geelong, Vic 3217, Australia
The hot strength of austenitic steels with the carbon content varying from 0.0037 to 0.79wt% was modeled using artificial neural networks. The carbon content has a complex effect on flow strength of austenite. An increase in carbon content reduces the flow stress of the steels at high temperatures and low strain rates, while it increases the flow stress at low temperatures and high strain rates, especially at low strains. In addition, increasing carbon to above 0.4 wt.% dramatically reduces the peak strain for the initiation of dynamic recrystallization at high Zener-Hollomon parameter, Z. Given the complexity of the deformation and recrystallization behaviors of these steels, no phenomenological or simple empirical models are able to predict the flow stress over the full carbon range. In this work, the back error propagation algorithm of the ANN model with one hidden layer bias was used, with the number of hidden nodes optimized. The data up to a strain of 4 were used to predict the strength in both work hardening and dynamic recrystallization regimes. The training speed was an important parameter and was optimized by trimming the data set and learning procedures. The effects of the carbon content on flow stress, peak strains and peak stresses observed from the experiment were accurately represented. However, it was found that the training data set also needed to be optimized to accurately predict the hot strength of the steels.
NOTE: This is just an abstract of the application summary. The entire summary is available at:
http://www.nd.com/application%20summaries/appsum-model.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 each newsletter, we’ll spotlight a new solution and include a link for people to get more information.
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Comments or Suggestions?
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
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!