Thermal Food Processing

108 Thermal Food Processing: New Technologies and Quality Issues
4.1 INTRODUCTION
Artificial neural networks (ANNs) are being successfully applied for a wide range
of problem domains in diverse areas, including engineering, physics, finance,
medicine, and others related to purposes of prediction, classification, or control.
This extensive success can be attributed to many factors:
1. Power of modeling — Neural networks are very sophisticated techniques
capable of modeling extremely complex functions. A priori
knowledge of the system is not needed for constructing the ANN
because the ANN will learn its internal representation from the input/output
data of its environment and response.
2. Ease of use — Neural networks learn by example. The user of neural
networks gathers representative data and then invokes training algorithms
to automatically learn the structure of the data. Although the
user does need to have some heuristic knowledge of how to select and
prepare data, how to select an appropriate neural network, and how to
interpret the results, the level of user knowledge needed to successfully
apply neural networks is much lower than that needed to use some
more traditional nonlinear statistical methods.
3. High computational speed — The ANN is an inherently parallel architecture.
The result comes from the collective behavior of a large number
of simple parallel processing units. Therefore, once trained, ANN can
calculate results from a given input very quickly. Because of this
feature, ANNs have a greater potential to be used for the online control
system than conventional modeling methods.
The concept of neural networks was based on the research in artificial intelligence,
which was specifically intended to mimic the fault tolerance and capacity
of biological neural systems by modeling the low-level structure of the brain.
Warren McCulloch and Walter Pitts 1 in 1943 were the first to open the idea on
how neurons might work, and they modeled a simple neural network using
electrical circuits. As computers became more advanced in the 1950s, it was
finally possible to simulate a hypothetical neural network. In 1959, Bernard
Widrow and Marcian Hoff developed models called ADALINE and MADA-
LINE. 2,3 In 1962, the same authors developed a learning procedure that examined
the value before the weight adjustment (i.e., 0 or 1), which was one of the
important fundamentals to the following success of neural networks. 4 However,
the neural network concepts did not result in practical applications until the 1980s,
when several new approaches, such as bidirectional lines, the hybrid network,
and multilayer neural networks, were developed. 2–6 In addition to these advances
in algorithms, the rapid development of computer technologies, including both
hardware and software, became an important driving force for neural networks
as a computing technique to be used not only in computing science, but also in
other areas as a tool for prediction, classification, and optimization.