EE 543 NEUROCOMPUTERS - BACKPROPAGATION ALGORITHM

Instructions:
Train MLP network to learn the selected patterns. In addition to predefined patterns, you may define also your own patterns.The network has 100 neurons since the pattern defined on 10x10 grid are to be trained to the network. The number of neurons at the output layer depends on the number of patterns that you have selected to be trained.In the network there are two hidden layers and you can adjust the number of neurons at each layer. Furthermore you can adjust the number of iterations, learning rate and momentum terms.You may observe the value of the error as training progresses.

After you have trained the network, load the patterns that you want on the grid, you may distort them by adding noise or by directly changing the elements on the grid by clicking them. Then ask the network to recognize the pattern. The charts at the top right of the applett are shows the value of each neuron output. In the ideal case the output of the neuron corresponding to the pattern will be 100% and all the others will be 0%.

Observe what happens as the amount of noiseis increased, and also as the number of iterations are increased.

Close this window