Page 307 - Machine Learning for Subsurface Characterization

Classification of sonic wave Chapter 9 269

FIG. 9.19 Artificial neural network with an input layer, one hidden layer, and an output layer.
ANN is fed six-dimensional feature vector. Hidden and output layers have five and one neurons,
respectively.

the immediately previous layer (input layer does not have a preceding layer).
A neuron is a mathematical transformation that sums all the inputs to the
neuron from the neurons of the previous layer multiplied by corresponding
weights of the corresponding connections between the neurons and then
applies a nonlinear filter or activation function (generally having a sigmoid
shape) on the summation to generate an output (Fig. 9.20). For example, if
the activation function of the output neuron is a sigmoid function ranging
from 0 to 1, the output will be in the range [0, 1]. The neurons in hidden
layers generally use the rectified linearunit(ReLU)ortanhas the activation

FIG. 9.20 An illustration of three hidden layers of an artificial neural network. Hidden layers #1,
#2, and #3 contain 2, 1, and 2 neurons, respectively. Two neurons (N 1 and N 2 ) in hidden layer #1 are
connected to four neurons in the previous layer. Outputs of the two neurons (N 1 and N 2 ) of hidden
layer #1 are fed to one neuron (N 1 ) in hidden layer #2. The single output of hidden layer #2 is fed to
the two neurons (N 1 and N 2 ) of the hidden layer #3. The inner working of the neuron (N 1 ) of hidden
layer #2 is presented for better explanation of the neuron as a computation unit.