A Computer Based Training (CBT) program on Neural Networks

Keith Ahern This project was undertaken in 1995 at the Wirtshaft Informatik (Business Informatics) department of the University of Dortmund in Germany. It was designed to be used both in the university and pressed on to a CD-ROM for use with a book on Neural Networks.

This interactive tutorial introduces the user to Neural Networks through the use of mulitmedia and interactive examples. The user is examined at the end of each chapter and an overall grade is given.

The course was authored in Asymetrix Multimedia Toolbook 3.0 and uses various types of media such as full-motion video, audio narrative, 3D graphics (3D Studio) and actual interactive Neural Networks. I was provided with (text) content and did all the programming, graphics (2D & 3D), animation and audio. There is the possibility of the course being published as a CD-ROM

The CBT was authored on a Pentium @ 60 Mhz with 16 megs of RAM running Windows 3.1. The target platform was a fast 386 or a 486 with VGA graphics.

Click on the image to see a full size version.

This is the contents page of the tutorial. Options available from here include a full search function, help window and an alternative contents page: "the brain". Every icon has 'tool tips' elaborating on its functions. The menus have graphical and audio feedback on mouseover events.

This is an example of an actual Neural Network in a real world situation. This credit evaluation system was trained on real world credit data, including whether the applicant was approved or disapproved. The user can enter in the data using the drop down boxes. When the data has been entered and the 'process' button has been pressed the Neural Net will activate and generate a result. The activation is shown visably by color changing the connections connectiing the neurons.

In this screenshot we see how the user is introduced to the non-linear activation function by using the analogy of a cars speed in proportion to its power. By clicking on the 'Mode' button the user flips the graph data from a car to a sigmoid function. The user can dynamically change parameters associated with the function and the results are shown in realtime e.g. if the power is increased messages are sent to the speed dial and graph which update them accordingly. The audio tutorial not only explains everything but also automatically animates the dials and graph in synchronisation with the audio. Needless to say, it is much more intuitive if you are using the CBT rather than looking at a static image.

This screenshot demonstrates the use of unsupervised training. The user is asked to separate the blocks in to two subsets, however they are not told HOW they should be separated. The user will split them up by color even though they were not explicited asked to, when they do this a paragraph is added to the screen explaining that they have just experienced unsupervised learning.