• Neural Network Course Completed

• TradingSolutions Real-Time Now Available!

• NeuroSolutions Tip Box:
  Training the Network Weights using Genetic Optimization

• Comments or Suggestions?

We successfully completed another neural network and NeuroSolutions course in early May. We always enjoy getting to know our customers better, especially when it gives us a chance to discuss their applications. Once again, we received excellent reviews from our attendees and have never had anyone tell us they would not recommend our course to others. In fact, a number of attendees were inquiring about coming back to future courses. As one attendee wrote – “Great Course -- I enjoyed it"

The next Neural Network course is being planned for November 2003. If you would like more information on our courses, please visit http://www.neurosolutions.com/products/course/

NeuroDimension is proud to announce version 2.1 of its AI-based technical analysis software, TradingSolutions. TradingSolutions now supports importing historical end-of-day and intraday data, as well as streaming live data from eSignal for use with intraday models.

TradingSolutions Real-Time automatically alerts the day trader of a new trading signal immediately after the model detects an opportunity for profit.

TradingSolutions End-of-Day includes many new enhancements as well, including asynchronous processing, improved tutorials, and numerous interface improvements.

TradingSolutions Real-Time is free to try for 30 days! Download the evaluation version from our website, check out the animated demos, learn from the step-by-step tutorials, and try it with any data you like for 30 days.

Download it today from: http://www.tradingsolutions.com/downloads/eval.html

Current customers of TradingSolutions v2.0 and previous versions can upgrade to TradingSolutions End-of-Day v2.1 for FREE! For additional upgrade and pricing information, please see the ordering information page on the TradingSolutions website at: http://www.tradingsolutions.com/order/

The Users level (and above) of NeuroSolutions includes Genetic Optimization. Genetic Optimization allows you to optimize virtually any parameter in a neural network to produce the lowest error. For example, the number of hidden units, the memory taps, the learning rates, and the input selection can all be optimized to improve the network performance.

What many NeuroSolutions users are not aware of is that Genetic Optimization can also be used to update the network weights as an alternative to backpropagation. The advantage is that the genetic algorithm can sometimes find a better solution than backpropagation because it is less prone to getting caught in a local minima. The big disadvantage is that the training is much slower than backpropagation since it is a less directed search, so it is generally only used when there are plenty of computing resources and the neural network has relatively few weights.

Below is an outline of the steps required to change a standard backpropagation breadboard to one that trains with genetic optimization instead:

1. Remove the backprop components by opening the BackStaticControl inspector (right-click on the red dials and select “Properties”) then click the “Remove” button.
2. Left-click on the orange dials to display the StaticControl inspector. Change the “Epochs/Run” to 1.
3. Close down all probes to speed training (optional).
4. Stamp a GeneticControl component (the green dials) on top of the BackStaticControl component (the red dials). Note: You may need to display the appropriate toolbar first by selecting “Customize” from the Tools menu then checking the “Control Components” checkbox.
5. Stamp on a DataGraph probe on top of the GeneticControl component and double-click the DataGraph icon to display the probe's window. The title should read “Best Fitness of controlGenetic”. If not, you need to right click on the GeneticControl icon, select Properties, switch to the “Access” tab and change the Access Point to “Best Fitness”.
6. Right click on the first FullSynapse component and select “Properties”. Switch to the “Genetic” tab and check the box entitled “weightValue”. Change the Lower Bound to -1.
7. Repeat step 6 for all remaining FullSynapses, as well as any TanhAxons, SigmoidAxons, LinearAxons or BiasAxons.
8. Reset and run the network to train it.
9. If you allow the network to run until it terminates on its own (100 generations by default) then the best weights will be loaded into the network automatically. If you stop the network early, then you will need to load the best weights by clicking the “Load Best Parameters” button from the GeneticControl inspector.
10. To test the network, uncheck the “Enable Optimization” switch from the GeneticControl inspector, then run the TestingWizard.

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: newsletters@nd.com.

Have questions about NeuroDimension products? Send your questions to: info@nd.com

This issue and previous issues of this newsletter are available on the NeuroSolutions web site at: http://www.neurosolutions.com/newsletters.html

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!