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
a.
Let
the biases on the
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
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)