Subject: NeuroDimension News and Tips V1 #4

Greetings from NeuroDimension!

Makers of NeuroSolutions, the Neural Network Simulation Environment.

NeuroDimension wants to help you to understand your data. This issue of the newsletter highlights the neural network courses, the online neural network reference section, and several topics to help you get the most out of your data.

In this issue you’ll find:

What’s New?

What’s Noteworthy?

Designing Neural Networks

NeuroSolutions Tip Box

Did You Know?

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 page for removal instructions.

What’s New?

Neural Network Reference Section

Performing research on neural networks? The NeuroDimension web site has a reference section where you can obtain information on current applications of neural networks. It contains the best neural network books, papers, and links for every neural network application from financial forecasting to medical diagnosis. Visit it today at http://www.nd.com/nnreference.htm

June Neural Network Courses Well-Received

NeuroDimension has completed another set of neural network training courses in Orlando. Thank you for everyone who attended. Our course feedback was again very positive with an average course rating of 4.6 on a scale of 1-5 (5 being best). We have yet to have an attendee say they would NOT recommend the courses to others. Look for our next Orlando course offering in the November to January time frame. For more information on NeuroDimension courses, see our course web page at http://www.nd.com/course

European Neural Network Course Survey

We are looking into offering a course in Europe. If you are interested, please fill out the short survey on our web site at http://www.nd.com/course/eurosurvey.htm. Our decision to offer the course will be based on the response to this survey.

What’s Noteworthy?

PCAI Application Articles

PCAI (http://www.primenet.com/pcai/) is an international magazine with a circulation of about 25,000 specializing in Artificial Intelligence for the PC. They are constantly looking for AI application developers to submit 2-4 page articles on their work. This is great advertising for your project or company. Please contact Gary Lynn (gary@nd.com) if you have a NeuroSolutions-based application that you would like to write about and have published.

Designing Neural Networks

This Month: Understanding Your Data

Neural networks are very powerful and with enough data and enough training they can approximate any input-output mapping arbitrarily well. There is no replacement, however, for understanding your data. Blindly applying neural networks to data will typically produce luke-warm results. By manipulating and preprocessing your data, however, you can usually increase performance and decrease training time dramatically.

For instance, in many control or prediction applications the derivative of the data is at least as important as the actual value of the data. The derivative contains the key information about which direction the data is moving (upward or downward slope). Computing the difference of successive inputs (an approximation to the derivative) and providing this additional information to the network may make the difference between good results and great results. Another example is frequency information. If you know that the peaks in the frequency spectrum are important to a specific classification problem, extract the peaks and feed that information to the network rather than every point in the frequency spectrum.

In summary, embed all the information you know about the problem into the input data through preprocessing. Remove any noise you know exists, remove outliers that may be obvious to you but not the neural network, and extract information that you already know is relevant. This allows the neural network to focus on information you don’t already know.

NeuroSolutions Tip Box

This Month: Sensitivity Analysis

After training a neural network, you may want to know the effect that each of the network inputs is having on the network output. The sensitivity analysis feature of NeuroSolutions can be used to perform this function.

Sensitivity analysis is a method for extracting the cause and effect relationship between the inputs and outputs of the network. The basic idea is that each input channel to the network is offset slightly and the corresponding change in the output(s) is reported. The input channels that produce low sensitivity values can be considered insignificant and can most often be removed from the network. This will reduce the size of the network, which in turn reduces the complexity and the training time. Furthermore, this may also improve the network performance.

For instructions on configuring your NeuroSolutions breadboard to report the sensitivity data, please perform a keyword search on "Sensitivity" within the on-line help. Note that NeuroSolutions for Excel also includes a (slightly different) sensitivity analysis function.

Did You Know?

This Month: Using a Custom Solution Wizard DLL within Microsoft Excel

Neural network DLLs generated using the Custom Solution Wizard can be accessed from Microsoft Excel by using the NeuroSolutions Object Library, an ActiveX component which is included with the Custom Solution Wizard. Using the NeuroSolutions Object Library makes communicating with the DLL much easier than calling it directly.

The protocol for using the NeuroSolutions Object Library is the same as it is for any other VB/VBA application, so instead of focussing on the protocol, this segment will focus on what is unique to Excel: the method for retrieving the data. Within Excel, data is contained in spreadsheets. The NeuroSolutions Object Library uses two properties to send the data to the network: inputData and desiredData. Both of these properties require the data to be Integer, Long, Single, or Double precision values wrapped by a Variant array. The dimensions of the input data must be (inputs x exemplars) and the dimensions of the desired data must be (outputs x exemplars). To get the data from an Excel worksheet into this format, you could create an array of the proper size then read the values one by one from the worksheet into the array. However, there is a much easier and much faster way to do this. Excel allows you to read data directly from a worksheet into an array.

For example, assume that your input data lies in the range A2:B5 and the desired data lies in the range C2:C5. You can use the following code to read these datasets directly into arrays:

Dim myInputData As Variant

myInputData = ActiveWorkbook.Sheets("MySheet").Range("A2:B5").Value

Dim myDesiredData As Variant

myDesiredData = ActiveWorkbook.Sheets("MySheet").Range("C2:C5").Value

This will read the data into Variant arrays in the form (exemplars x inputs) and (exemplars x outputs), respectively, so you will have to transpose the arrays before assigning them to inputData or desiredData.

For large datasets, using this method for retrieving the data can really make a difference (up to 10 times faster). See the "inputData Property" or "desiredData Property" topics in the Custom Solution Wizard on-line help for 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!