HOW TO SUBMIT HOMEWORKS

Assignment#1:LOADING AND EDITING MODULE : DUE TO NOVEMBER 23, 2005

 

• It should display two different versions of patterns for each of the numerals 0,1,2,3,4,5,6,7,8,9 on 10x8 grids (10 rows, 8 columns)
• Each pattern should have a check box to mark the patterns selected for training which will be functional in assignments #2 and #3
• There should be buttons for training and recall (classify) operations which will be functional in assignments #2 and #3
• It should be possible to select a pattern for loading by a mouse click on it
• There should be a larger 10x8 grid to display the loaded pattern
• It should be possible to edit the loaded pattern
• It should be possible to add noise to the loaded pattern
• There should be a help button to explain the operation of the applett
• You may use any additional item that you feel necessary
• On your applet canvas reserve some empty area considering assignments #2 and #3 for not having difficulty later

 

Assignment#2:  AUTOASSOCIATIVE MEMORY: DUE TO DECEMBER 7, 2005

• Implement synchronous Hopfield network functioning as autoassociative memory
• The patterns are to be the same as defined in assignment#1
• The patterns are to be loaded, edited and selected for training as defined in assignment#1
• Activate buttons for training and recall operations
• In addition to the large grid for loading and editing the selected pattern as implemented in assignment#1, there should be another large 10x8 grid to display the recalled pattern
• During recall, display not only the final state but also the intermediate states of the network one after another on the grid reserved for recall
• You may use any additional items that you feel necessary

Assignment#3: MULTILAYER PERCEPTRON BY BACK PROPAGATION DUE TO DECEMBER 28, 2005

 

• Implement Multilayer Perceptron having tanh as output function trained by Backpropagation Algorithm for pattern classification
• The patterns are to be the same as defined in assignment#1 (you have two samples for each class)
• The patterns are to be loaded, edited and selected for training as as defined in assignment#1
• Activate train and classify buttons
• It should be possible to set the number of neurons in the hidden layers but being assigned default values that you find appropriate
• Notice that the number of neurons at the input layer should be 80 which is the size of the patterns to be trained. The number of neurons at the output layer should be the same as the number of classes. In our case the classes corespond to numerals selected for training, each having two samples. In the ideal case, the output neuron representing the class of the applied pattern is expected to have output value 1 while all the others have -1.
• Display the the output of the neurons at output layer as a bar chart extending from -1 to +1 together with its numerical value
• There should be help button to explain the operation of the applet
• You may use any additional items that you feel necessary