Sunteți pe pagina 1din 4

1111111111111I11I111111mlllllllllllllll~ Il!

:ill~11111111111 ETC - 7(E-II) 2(RC)

B.E. (ETC) (Semester - VII) (RC) Examination, Nov./Dec. 2013


ARTIFICIAL NEURAL NETWORK

Duration: 3 Hours Total Marks: 100

Instructions: i) Assume suitable data if necessary.


ii) All symbols carry their usual meanings.
iii) Answer any five questions, selecting at least one question
from each Module.

MODULE-I

1. a) Explain how Mc-Culloch Pitt's neuron model is derived from biological neuron. 4
b) Design the neural network for the printed characters L and I in 3x3 pixel
window. Assign input vector entries 0 to white and 1 to black pixel respectively.
What are disadvantages of ANN? 8
c) A diachotomiser has to be trained to assign following points to the class as
mentioned below:
X1 = [-
2], X2 = 2/3] ~ class I r
X3 = [1], X4 = [2] ~ class II.
i) Draw separating lines for each pattern in augmented weight space.
ii) With wt
= [1, 1] display the movement of weight vector on weight plane
for 1 iteration using bipolar sigmoid activation function. 8

2. a) State and explain co-relation learning rule and Widraw-Hoft learning rule. 6
b) For the following network use X = 1.8, unipolar sigmoid activation function
and findout which neuron fires for following inputs.

x = [0.866] X = [- 0.985] X = [0.342]


1 0.5 2 -0.174 3 -0.42

[1;1 ~_-_O.....::'-9:....;:6-=6_--I
r---Ol

P.T.O.
ETC - 7(E-II) 2(RC) ·2· 1111111111111111111111111I1111111111111111111111111111111111

c) Design the neural network for following classification problem:

E§" - Class 'r


~ - clCJ~n
o - Ci(J~JII
8

MODULE-II

3. a) On what the speed of error back propagation algorithm depends? How to


overcome the same? 8
b) Explain the different ways to get a optimum sized network. 6
c) Write a short note on generalizability. 6

4. a) Apply Marchand's algorithm for corner isolation problem. 10

b) Why radial basis function is required? Define radial basis function. What are
different radial basis functions?
c) Write a short note on polynomial networks. 4

MODULE -III

5. a) Cluster the following points using simple competitive learning algorithm.


Assume n = 0.5
X1 = [0 1 1] X4 = [0.5 0.5 0.5]
X2 = [1 1 0.5] Xs = [0.4 0.6 0.5]
X3 = [0.2 0.2 0.2] X6 = [0 0 0]
1111111111I11111111111111I111111111111111111111111111 m 1111 -3- ETC - 7(E-II)2(RC)

Assume initial weight matrix as

l
O.14 0.75 0.71j
W = 0.99 0.51 0.37
12
0.73 0.81 0.87

b) Explain LVQ algorithm. What is the necessity of LVQ algorithm? How it is


different from simple competitive learning algorithm? 8

6. a) Apply ART to following points with one complete iteration. Assume vigilance
parameter = 0.7
(1010001)
(00 1 1 1 1 0)
(1100001)
(000 1 1 1 1) 8
b) Apply SOFM algorithm for patterns shown below
X1 = [1.5, 2.8, 4.6], X2 = [1.4, 3, 4.2]
X3 = [1.~, 3.2, - 3.9], X4 = [1.2, 2.9, 4.7]
The weights with three node structure are arranged as@-®-©Nith
neighbourhood distance 1 and n = 0.5. Assume connection weights as

l
O.2 0.7 0.1j
W = 0.1 0.1 0.4
111
8
c) Explain anyone type of distance based learning. 4
MODULE-IV
7. a) What are Hopefield networks? What are different types of Hopefield networks?
What are different methods of weight updation of Hopefield network? 8
b) Design hetro associative memory using matrix associative memory technique
for the following input output set
liP O/P
(1,1,-1) >(-1,-1)
(1,-1,-1) >(1, -1)
(-1,-1,1) >(1,1)
(-1,1,1) >(-1,1) 8
Ere - 7(E-II) 2(RC) -4- 1111111111111111111111111I1111I111111111111111I1111111111111

c) Consider four node Hopefield network. Whose weights store patterns


(1, 1, 1, 1), (- 1, - 1, -1, - 1) ? Determine the network response for
(0,0,0, -1). 4

8. a) What is BSB network? Where it is used? Explain the working principle of the
same. 6
b) Design the BSB network to store the patterns
(1, 1, 1), (- 1, -1, - 1), (-1, - 1, 1). Determine network response for corrupted
input pattern (0.5, 0.6, - 0.7). 6
c) Explain anyone application of neural network. 8

S-ar putea să vă placă și