• About Me
• Curriculum Vitae
• Interaction Design
  • Projects »
    • Interactive Canvas
    • Balloon Wars
    • Bus Display
    • Dating Booth
    • Do Avatars Dream?
    • Drawing Turtle
    • Life Timer
  • What's Interaction Design?
• Programming
  • 3D World Editor
  • Actionscript 3
  • A.I. Algorithms »
    • Flocking Algorithm
    • Genetic Algorithm
    • Neural Networks
  • Serial Comm. Tools
  • SMS Service
• Blog
• Contact

Neural Networks

Please refer to the sites below for an introduction into Artificial Neural Networks (ANNs) and their uses:

www.ai-junkie.com
www.doc.ic.ac.uk [Imperial College report]

My Neural Net Classes

I have programmed two sets of Neural Net classes that simulate an Adaline and a Perceptron Net respectively. You can download the project files (which compile as console applications under Visual C++ 6) here:

Download - adaline.zip (file size: 55 KB)
Download - perceptron.zip (file size: 142 KB)

The Adaline:

The Adaline project (.dsp) file contains the following source files (click on the links to view the code):

• adaline.h - adaline class declarations
• adaline.cpp - class definitions
• testAdaline.cpp - test program using the adaline classes

An ADALINE (ADAptive LInear NEuron) consists of a single processing unit. It is a very basic form of artificial neuron; you can't connect them together to build networks as you would with a Perceptron. I wrote the code purely for learning purposes before progressing on to the Perceptron Net.

The Multi-Layer Perceptron (MLP) Networks:

The Perceptron project (.dsp) file contains the following source files (click on the links to view the code):

• perceptron.h - MLP Net class declarations
• perceptron.cpp - class definitions
• testPercept.cpp - test program using the MLP Net classes

I have produced a diagram which illustrates an example Perceptron Network that can be configured using my classes.
Click HERE to view the diagram.

Multi-Layer Perceptron networks are probably the best know and most widely used of all Artificial Neural Networks. They are often used to tackle problems such as hand-writing or voice recognition. An MLP net is constructed by arranging perceptron units together in a series of 'layers'. The neurons in each layer are connected to the neurons in the adjacent layers in a hierarchical fashion (the neurons outputs from one layer supply the inputs for the next). Typically, each neuron within a layer will have an input 'dendrite' connected to the output of every neuron in the previous layer (this is known as a FULLY-CONNECTED network, which is the method employed in my classes). The bottom layer is known as an 'input layer' and usually receives input data from sensors or some other input device(s). Further documentation of my perceptron classes can be found here: neural.pdf.

Top of page