Training the SOM

• Training set:
  • Contains inputs only
  • No labels needed
• Training parameters
  • Number of iterations
    • Each iteration applies each input to the network
  • Learning rate
    • Between 0 and 1
    • Starts at 1; decreases with each iteration
  • Radius
    • All output nodes within the radius of the winner are trained
    • All other nodes are ignored
    • Starts large; decreases with each iteration