3. Artificial Neural Network: Structure
• ANN posses a large number of processing elements called
nodes/neurons which operate in parallel.
• Neurons are connected with others by connection link.
• Each link is associated with weights which contain
information about the input signal.
• Each neuron has an internal state of its own which is a
function of the inputs that neuron receives- Activation level
3Neural Networks Dr. Randa Elanwar
4. Artificial Neural Network: Neuron Model
(dendrite) (axon)
(soma)
4Neural Networks Dr. Randa Elanwar
f()
Y
Wa
Wb
Wc
Connection
weights
Summing
function
Computation
(Activation Function)
X1
X3
X2
Input units
5. How are neural networks being used in
solving problems
• From experience: examples / training data
• Strength of connection between the neurons is
stored as a weight-value for the specific connection.
• Learning the solution to a problem = changing the
connection weights
5Neural Networks Dr. Randa Elanwar
6. How are neural networks being used in
solving problems
• The problem variables are mainly: inputs, weights and
outputs
• Examples (training data) represent a solved problem. i.e. Both
the inputs and outputs are known
• Thus, by certain learning algorithm we can adapt/adjust the
NN weights using the known inputs and outputs of training
data
• For a new problem, we now have the inputs and the weights,
therefore, we can easily get the outputs.
6Neural Networks Dr. Randa Elanwar
7. How NN learns a task: Issues to be
discussed
- Initializing the weights.
- Use of a learning algorithm.
- Set of training examples.
- Encode the examples as inputs.
-Convert output into meaningful results.
7Neural Networks Dr. Randa Elanwar
8. Linear Problems
• The simplest type of problems are the linear problems.
• Why ‘linear’? Because we can model the problem by a
straight line equation (ax+by+c=z)
• or
• Example: logic linear problems And, OR, NOT problems. We
know the truth tables thus we have examples and we can
model the operation using a neuron
8Neural Networks Dr. Randa Elanwar
bout
k
i
ii inw 1
.
outbinwinwinw ...... 332211
bXWOUT .
9. Linear Problems
• Example: AND (x1,x2), f(net) = 1 if net>1 and 0 otherwise
• Check the truth table: y = f(x1+x2)
9Neural Networks Dr. Randa Elanwar
x1 x2 y
0 0 0
0 1 0
1 0 0
1 1 1
x1
x2
y
1
1
10. Linear Problems
• Example: OR(x1,x2), f(net) = 1 if net>1 and 0 otherwise
• Check the truth table: y = f(2.x1+2.x2)
10Neural Networks Dr. Randa Elanwar
x1 x2 y
0 0 0
0 1 1
1 0 1
1 1 1
x1
x2
y
2
2
11. Linear Problems
• Example: NOT(x1), f(net) = 1 if net>1 and 0 otherwise
• Check the truth table: y = f(-1.x1+2)
11Neural Networks Dr. Randa Elanwar
x1 y
0 1
1 0
x1
y
-1
2
bias
1
12. Linear Problems
• Example: AND (x1,NOT(x2)), f(net) = 1 if net>1 and 0
otherwise
• Check the truth table: y = f(2.x1-x2)
12Neural Networks Dr. Randa Elanwar
x1 x2 y
0 0 0
0 1 0
1 0 1
1 1 0
x1
x2
y
2
-1
13. Neural Networks Dr. Randa Elanwar 13
The McCulloch-Pitts Neuron
• This vastly simplified model of real neurons is also known as a Threshold Logic Unit
– A set of connections brings in activations from other neurons.
– A processing unit sums the inputs, and then applies a non-linear activation function (i.e.
squashing/transfer/threshold function).
– An output line transmits the result to other neurons.
).(
1
bfout
n
i
ii inw
f(.)
w1
w2
wn
b
).( bXWfOUT
15. Features of McCulloch-Pitts model
• Allows binary 0,1 states only
• Operates under a discrete-time assumption
• Weights and the neurons’ thresholds are fixed in the model
and no interaction among network neurons
• Just a primitive model
Neural Networks Dr. Randa Elanwar 15
16. McCulloch-Pitts Neuron Model
• When T = 1 and w = 1
• The input passes as is
• Thus if input is =1 then o = 1
• Thus if input is =0 then o = 0 (buffer)
• Works as ‘1’ detector
• When T = 1 and w = -1
• The input is inverted
• Thus if input is =0 then o = 0
• Thus if input is =1 then o = 0
• useless
16Neural Networks Dr. Randa Elanwar
17. McCulloch-Pitts Neuron Model
• When T = 0 and w = 1
• The input passes as is
• Thus if input is =0 then o = 1
• Thus if input is =1 then o = 1
• useless
• When T = 0 and w = -1
• The input is inverted
• Thus if input is =1 then o = 0
• Thus if input is =0 then o = 1 (inverter)
• Works as Null detector
17Neural Networks Dr. Randa Elanwar
18. McCulloch-Pitts NOR
18Neural Networks Dr. Randa Elanwar
•Can be implemented using an OR
gate design followed by inverter
•We need ‘1’ detector, thus first layer
is (T=1) node preceded by +1 weights
Zeros stay 0 and Ones stay 1
•We need inverter in the second
layer, (T=0) node preceded by -1
weights
•Check the truth table
19. McCulloch-Pitts NAND
19Neural Networks Dr. Randa Elanwar
•Can be implemented using an
inverter design followed by OR gate
•We need inverter in the first layer is
(T=0) node preceded by -1 weights
Zeros will be 1 and Ones will be
zeros
•We need ‘1’ detector, thus first layer is
(T=1) node preceded by +1 weights
Zeros stay 0 and Ones stay 1
20. General symbol of neuron consisting of
processing node and synaptic connections
Neural Networks Dr. Randa Elanwar 20
21. Neuron Modeling for ANN
Neural Networks Dr. Randa Elanwar 21
Is referred to activation function. Domain is
set of activation values net. (Not a single
value fixed threshold)
Scalar product of weight and input vector
Neuron as a processing node performs the operation of summation of its
weighted input.