Virtual Lab
  • Dashboard
  • Perceptron Learning
    • Objective
    • Tutorial
    • Illustration
    • Procedure
    • Simulation
    • Assignment
  • Backpropogation Network
    • Objective
    • Tutorial
    • Illustration
    • Procedure
    • Simulation
    • Assignment
  • Multilayer Perceptron
    • Objective
    • Tutorial
    • Illustration
    • Procedure
    • Simulation
    • Assignment
  • Kohonen Self-sOrganizing Net
    • Objective
    • Tutorial
    • Illustration
    • Procedure
    • Simulation
    • Assignment
  • Learning Vector Quantization
    • Objective
    • Tutorial
    • Illustration
    • Procedure
    • Simulation
    • Assignment

Algorithm

Step 1: Initialization

Set initial synaptic weights to small random values say in interval [0,1] and then assign a small positive value to learning rate parameter α

Step 2 : Activation and Similarity Matching

Activate Kohonen network by applying input vector X and find winner takes all (best match neuron j(x) at iteration p, using minimum distance Euclidean Criterion.
jx(p) = min ||X- wj(p) || ={∑ [x(i) - w(ij)(p)]^2}^1/2 where n is no of neurons in input layer,
and m is no of neurons in Kohonen layer

Step 3: Learning

Update synaptic weights
w(ij)(p+1) = w(ij)(p) + Δw(ij)(p)
where Δw(ij)(p) is weight correction at iteration p
Weight correction is determined by competitive learning
Δw(ij)(p) = α [x(i) - w(ij)(p)] if j ∈ ^ j(p)
Δw(ij)(p) = 0 if j ∉ ^ j(p)
where α is learning rate parameter and n(j)(p) is neighbourhood function centered around winner takes all neuron j(x) at iteration p