Geology Reference
In-Depth Information
Fig. 4.3 a Topology of a
three-layer feed-forward
neural network. b Detailed
view of connections from a
node
layer node p or network input p, w
c,p
is the weight modifying the connection from
either node p to node c or from input p to node c, and b
c
is the bias. The subscripts
c, p, and n in the given equations in this section will identify nodes in the current
layer, the previous layer, and the next layer, respectively. The sigmoid activation
function was employed in this research, which is shown in Fig.
4.4
.In(
4.17
),
h
Hidden
(x) is the sigmoid activation function of the node. Before performing the
training process, the weights and biases are initialized to appropriately scaled
values. Appropriate normalization of training data is essential to avoid saturating
the activation function.
For output layer linear activation a function was used. Thus, the output layer
nodes perform the calculation as follows:
!
h
Output
X
P
O
c
¼
i
c
;
p
w
c
;
p
þ
b
c
where h
Output
ð
x
Þ
¼
x
ð
4
:
18
Þ
p¼1
