HOW TO SUBMIT HOMEWORKS

PROGRAMMING ASSIGNMENTS -2008 (last updated 25-09-2008)

Assignment#1: MULTILAYER PERCEPTRON: DUE TO OCTOBER 20, 2008

 

  • Display an 2 input, 1 output NN in multilayer structure with a single hidden layer having 4 neurons
  • There should be a separate scroll bar for each input (similar to those in sample neuron applet) in the range -10 to +10
  • There should be a seperate scroll bar for each of the connections from inputs to hidden neurons and also for the thresholds of these hidden neurons in the range -10 to +10
  • Assume that all the connections from the input layer to the output layer are fixed and has value 1, and the threshold value or the output neuron is -3, which is corresponding to AND operation
  • Assume all the hidden neurons and the output neuron has sign function as their output function which is has value -1 if the activation is negative, 0 if the activation is zero and +1 if the activation is positive
  • Show the input space u1, u2 as a rectengular area ranging from -10 to +10 in both dimensions. Indicate the current input values by a marker in the input space area. This marker should move as the input values u1 and u2 are changed through the scroll bars.
  • On this input space show the lines (similar to one given in the perceptron applet) represented by each hidden neuron i, i=1..4, ( w1i * u1+w2i * u2 + theta i =0) by a different color. These lines are to be shift and rotated as the connnection weights and thresholds are changed through the scroll bars.
  • On the input space, show regions for which the output neron has value -1 or +1 by coloring the pixels by black and white.
  • Additionaly, the output value should be displayed explicitly as the curser moved on the input space
  • Provide help on the usage of your applet which can be accessed through a help button
  • You may use any additional items that you feel necessary
  • Read also the assignments #2 and #3 befor starting to code this one. I suggest you to use a color image to represent the input space, to make the changes on this image piksel by piksel and to refresh the image on the screen after all the piksel values are set. However, if you find other ways that provide solution for whatever requested, they are also acceptable.

 

Assignment#2:  BACKPROPAGATION LEARNING IN MULTILAYER PERCEPTRON: DUE TO DECEMBER 1, 2008

  • Implement Backpropagation algorithm to learn two classes on two dimensional input space
  • Your network will have the same structure in the assignment #1, (that is 2 input neurons, 4 hidden neurons and 1 output neuron), but this time all the weights an the thresholds are to be learned through the backpropagation algorithm
  • The hidden and output neurons will have tanh function as the output function.
  • Still the scroll bars for the inputs should remain to change the input values
  • As in the assignment #1, show the input space u1, u2 as a rectengular area ranging from -10 to +10 in both dimensions. Indicate the current input values by a marker in the input space area. This marker should move as the input values u1 and u2 are changed through the scroll bars.
  • The samples for the training set are to be constructed by the user by mause click on the input space (in a way similar to the perceptron applet). The output value in the training set is to be set to +1 for the samples belonging to class 1, and it is set to -1 for the samples belonging to class 2
  • Once the training set is constructed, your network should learn the samples in the training set when the train button is pressed.
  • Once the training is completed, show the lines represented by each hidden neuron on the input space.
  • On the input space, indicate the network output value by coloring each piksel by a grey level values ranging from 0 to 255.
  • Additionaly, the output value should be displayed explicitly as the curser moved on the input space
  • Provide help on the usage of your applet which can be accessed through a help button
  • You may use any additional items that you feel necessary

Assignment#3:RADIAL BASIS NETWORK DUE TO DECEMBER 29, 2008
    • Implement a Radial Basis Function network having the same structure given in the assignment #1 (that is 2 input neurons, 4 hidden neurons and 1 output neuron), but this time also the weights and the threshold of the output layer is to be adjusted through scroll bars.
    • Assume the hidden neurons has Gaussian output functions whose centers in two dimentionl space are represented by their connection weights from the input layer.
    • The variance of the Gaussians are all to be the same and its value is to be entered by the user through a text box
    • As in the assignments #1 and #2, show the input space u1, u2 as a rectengular area ranging from -10 to +10 in both dimensions. Indicate the current input values by a marker in the input space area. This marker should move as the input values u1 and u2 are changed through the scroll bars.
    • On the input space, indicate the network output value by coloring each piksel by a grey level values ranging from 0 to 255.
    • Additionaly, the output value should be displayed explicitly as the curser moved on the input space
    • Provide help on the usage of your applet which can be accessed through a help button
    • You may use any additional items that you feel necessary.