home *** CD-ROM | disk | FTP | other *** search
- THE VISIBLE NEURAL NETWORK VERSION 2.0
-
- "The Visible Neural Network," Version 2.0 is the first complete
- neural network package available in shareware. For your
- registration fee of $25 you will receive a fully illustrated spiral
- bound manual which completely describes the feed-forward / back-
- propagation algorithm in terms any non-mathematician (or
- mathematician) can easily understand, source code listings for
- both the tutorial program and OCR demo, and detailed instructions
- for using the application environment.
-
- If you want to become a leading expert on neural networks, "The
- Visible Neural Network" is a great bargain. Register by sending
- $25 to:
-
- ╔═════════════════════════════════════════════════════╗
- ║ ║
- ║ The Visible Neural Network, Inc ║
- ║ 4228 Weskan Court ║
- ║ Bridgeton, MO 63044-1317 ║
- ║ ║
- ║ (314) 739-2654 (MC/Visa accepted 24 hours) ║
- ║ ║
- ╚═════════════════════════════════════════════════════╝
-
- IN ORDER TO WRITE EFFECTIVE NEURAL NETWORK APPLICATIONS, YOU NEED
- THE KIND OF THOROUGH UNDERSTANDING OF THE ALGORITHM THAT THE
- REGISTERED PACKAGE CAN GIVE YOU. REGISTER TODAY!!!
-
- The first thing that you should do in using this package is to run
- the ocr.exe file,"Optical Character Recognition Demo." Optical
- Character Recognition (OCR) is one of the most impressive
- demonstrations of the ability of an artificial (non-biological)
- neural network to learn. Neural networks learn using the same
- mathematical algorithm as the nerves of your brain (which is a
- biological neural network). The algorithm was determined by
- measuring the electrical activity of the neurons of a certain
- species of sea slug with particularly large nerve cells, and by the
- efforts of a number of researchers to interpret the results.
-
- Watch the OCR network train for awhile. Characters are being
- "scanned" by the input neurons in the scanner grid. The neurons
- that are "touched" by a character are excited and transmit signals
- through the network. Over many iterations of the character set,
- the network learns to activate the output neuron that we have
- designated as "A" when the input pattern "A" is scanned, and the
- same for "B" and "C" and all the rest of the input / output sets of
- patterns.
-
- Once you believe that neural network actually learn, the rest is
- relatively easy. There are only three equations involved, and most
- of the mathematics is just multiplication and addition. "The
- Visible Neural Network" tutorial program "visible2.exe" is a sort
- of neural network laboratory where you can observe a small neural
- network in action and stop it whenever you like to see what is
- going on inside of it. Run the program and explore the menu
- selections. When you understand "Numerical Analysis" screens 4
- through 7 you will be a leading expert on neural networks!
-
- Because of their ability to recognize patterns, neural networks are
- reportedly being used to forecast results based on past experience,
- such as stock market analysis and gambling on sporting events.
- "The Visible Neural Network Application Environment" makes it easy
- to create, train and use neural network applications using data
- stored in NAMED RANGES of Lotus 1-2-3 *.WKS (not *.WK1 files). The
- data is automatically normalized for network input / output.
-
- THE SHAREWARE LOTUS-COMPATIBLE SPREADSHEET PROGRAM "ASEASYAS" HAS
- A TENDENCY TO PLANT PHANTOM RANGES IN WORKSHEETS AND IS NOT
- RECOMMENDED FOR USE WITH THE APPLICATION ENVIRONMENT.
-
- Running the application environment requires the following files:
-
- start.exe
- center.exe
- create.exe
- train.exe
- use.exe
- large.exe
- read123.exe
- write123.exe
- netdata.wks
- testdata.wks
-
- The "netdata.wks" file is an example of how to set up your network
- training data. Load this file into your spreadsheet progam and
- look at the "INPUTDATA" and "OUTPUTDATA" ranges. When you create
- applications of your own you should put the input / output training
- data in ranges named "INPUTDATA" and "OUTPUTDATA" in the same way.
-
- The "testdata.wks" file is an example of how to set up data that
- you want to feed through a trained network. Load the file into
- your spreadsheet program and note the "TESTDATA" and "RESULTS"
- ranges. Set up your test data files the same way.
-
- To use the application environment, enter the word "start." Select
- the "Create Network File" menu choice and enter the name that you
- want to give the training file (such as "NETFILE"). Next enter the
- name of the training data worksheet ("NETDATA" in the case of our
- example). The program will read the spreadsheet ranges and ask you
- how many middle layer neurons you want with a suggested number in
- parantheses. Enter any number you please, but the suggested number
- is probably the minimum that will allow the network to converge
- upon a solution set.
-
- Now that you have created a network, train it by selecting the
- "Train Network" menu choice and entering "NETFILE." Set the
- training tolerance to the default limit of 0.10 and randomize the
- weights if you wish. Allow the network to train until the training
- tolerance is reached, at which point the network is fully
- "trained."
-
- (If you get an "out of memory" error when you attempt to train a
- network, try training it with the stand-alone program "large.exe."
- This program will accomodate much larger neural networks because it
- uses dynamic arrays instead of static arrays, but is slower because
- dynamic arrays use pointers to pointers and far memory segments.)
-
- Use the trained network by selecting the "Use Network" menu option.
- Enter the name of the trained network file "NETFILE" and the name
- of the test data worksheet "TESTDATA." The network stores the
- feed-forward data and results in a spreadsheet file "RESULTS.WKS"
- in ranges named "TESTDATA" and "RESULTS." View this file with your
- spreadsheet program.
-
- THE TEST DATA SPREADSHEET FILE MUST HAVE BOTH A "TESTDATA" AND
- "RESULTS" RANGE EVEN THOUGH THE "RESULTS" RANGE IS EMPTY. THE
- PROGRAM USES THIS FILE AS A TEMPLATE FOR THE "RESULTS.WKS" FILE.
-
- For help using this package, call (314) 739-2654. We will be happy
- to return your call COLLECT if you leave a message.
-
-
-
-
- GOOD LUCK!!!
-
-