Page 23 - Biosystems Engineering

4 Cha pte r O n e

flocking birds (or schooling fish or swarming insects). SVMs are
learning kernel-based systems that use a hypothesis space of linear
functions in high-dimensional feature spaces. Hybrid systems use a
mixture of two or more machine learning methods to take advantage
of their collective features. Machine learning methods have been used
in various engineering and scientific applications to extract knowl-
edge from high-dimensional and complex data and to solve optimi-
zation problems.

1.2.1 Neural Networks
Neural networks generally consist of a number of interconnected
processing elements known as neurons. The way neurons are inter-
connected or how interneuron connections are arranged determine the
architecture of a neural network. The method by which the strengths
of the connections (known as weights or synaptic weights) are adjusted
or trained to achieve a desired overall behavior of the network is gov-
erned by the learning algorithm used. NNs are classified according to
their architecture and learning algorithms.
The most popular architecture is a feedforward neural network,
where the neurons are grouped into layers. All connections are feed-
forward; that is, they allow information transfer only from an earlier
layer to the next consecutive layers. Neurons within a layer are not
connected, and neurons in nonadjacent layers are not connected.
Input signals are presented to the network via an “input layer.” The
nodes in the input layers do not process input signals but pass them
to one or more “hidden layers” where the actual processing is done
via a system of weighted “connections.” The hidden layers then link
to an “output layer” that provides the outputs of the network.
Figure 1.1 depicts a multilayer feedforward neural network that has
four layers: an input layer, two hidden layers, and an output layer. As

Input layer Hidden layer Output layer
q 1
w 11 O 1
x 1
w 12
w 1n O 2 q 2
y
x 2

O I q k
x n

FIGURE 1.1 Multilayer feedforward neural network.