NSF/REU Homework Neural Networks

advertisement
REU Site Program in CVIP
Summer 2005
Lab 7 – Exercises on Neural Networks
1) Consider the decimal digits 0-9 constructed using a 7 segment display. The seven
segments begin at the top (horizontal) and move around the display, with the middle
segment being the last.
1
|
6|
|
|
|
5|
|
|
7
|
|2
|
|
|
|3
|
|
4
A digit is represented as a binary string. Thus
0000000
nothing – not legal
1111110
0
0110000
1
1101101
2
1111001
3
etc.
a.) Using only a single-layer perceptron network, can you devise a scheme so that
only two output values (a single output node) are given. A 1 if the number (binary
string) is even or a 0 if the number is odd.
b.) Using a two-layer backpropagation network, can you devise a scheme so that the
single output is the decimal equivalent of the binary input?
c.) Using a radial basis function network, can you devise a scheme so that the single
output is the decimal equivalent of the input?
2) Consider the following function
f ( x1 , x2 )  5.5( x1  x22 )  2.2 x13  log10 x2
 1  x1  2, 1  x2  1000
a.) Build a backpropagation network to implement this function.
b.) Build a RBF network to implement this function.


To solve this problem you will need to generate train, verify, and test data.
For both problems (1 & 2), e-mail to the TA a matlab file containing your
network, and a file containing what you have found out about these problems.
Download