Using MATLAB tools for MLP NNs (e.g., newff, …), design a two-layer feed-forward neural network as a classifier to categorize the input geometric shapes.

- The snapshot and bitmap of shapes are given:

- Training shapes: shkt.bmp

- Training patterns: trn.txt (each shape is in a 125*140 matrix)

- Test shapes: shks.bmp

- Test patterns: tsn.txt (each shape is in a 125*140 matrix)

- Since the dimension of inputs is too high (17500-dimensional), it is not possible to apply them directly to the net. So, … .

- Try the number of hidden neurons to be at least.

- Do training of NN until all training patterns are truly classified.

- To examine the generalization ability of your NN after training,

a) Apply it to the test patterns and report the accuracies.

b) Add p noise (p=5, 10, …, 75) to the training shapes (only degrade the

black pixels of the shapes) and report in a plot the accuracy versus p.-Using MATLAB tools for MLP NNs (e.g., newff, …), design a two-layer feed-forward neural network as a classifier to categorize the input geometric shapes.

- The snapshot and bitmap of shapes are given:

- Training shapes: shkt.bmp

- Training patterns: trn.txt (each shape is in a 125*140 matrix)

- Test shapes: shks.bmp

- Test patterns: tsn.txt (each shape is in a 125*140 matrix)

- Since the dimension of inputs is too high (17500-dimensional), it is not possible to apply them directly to the net. So, … .

- Try the number of hidden neurons to be at least.

- Do training of NN until all training patterns are truly classified.

- To examine the generalization ability of your NN after training,

a) Apply it to the test patterns and report the accuracies.

b) Add p　noise (p=5, 10, …, 75) to the training shapes (only degrade the

black pixels of the shapes) and report in a plot the accuracy　versus p.