Sample Questions from Week 5 – Backpropagation and Supervised learning

 

1.        Network Design and training

a.        The truth table for the 3-bit AND function is as follows:

X

Y

Z

AND

0

0

0

0

0

0

1

0

0

1

0

0

0

1

1

0

1

0

0

0

1

0

1

0

1

1

0

0

1

1

1

1

 

1.        Draw a neural network that could learn the AND of three inputs, x, y, and z.

2.        how may input, hidden and output units would your design need for a minimal design?

3.        what learning algorithm would be appropriate?

 

b.       Fill in the following truth table for the parity function (Parity is zero if the sum of the inputs is even, and one if the sum is odd).

X

Y

Z

Parity

0

0

0

 

0

0

1

 

0

1

0

 

0

1

1

 

1

0

0

 

1

0

1

 

1

1

0

 

1

1

1

 

1.        Draw a neural network that could learn the parity function of three inputs, x, y, and z. 

2.        how may input, hidden and output units would your design need for a minimal design?

3.        what learning algorithm would be appropriate?

 

2.        The 1-1-1 network

a.        Let the biases on the 1-1-1 network be zero, and w1=0 and w2=-8.

1.        what would the TSS be (use the table from the lecture last week)?

2.        track the weight changes and the TSS through the table and on the 1-1-1 error surface

3.        Track the changes from the initial point, w1=w2=1.

4.        what happens at the point w1=w2=0?

 

3.        What are the practical consequences in a backpropagation network if the learning rate is

a.        moderately large (e.g., lrate=0.5)

b.       very large (e.g., lrate=10)

c.        very small (e.g., lrate=0.0001)