Neural computing in pharmaceutical products and process development - Computer-Aided Applications in Pharmaceutical Technology - page 92

5.1 Artifi cial neural networks

5.1.1 Introduction

The way that the human nervous system is organized and functions
has always attracted great attention. Principles of brain functions,
such as learning and memory, were studied by both medical and
computer scientists. The laws underlying human thinking and reasoning
have been the basis for machine learning algorithms. It was, reasonably,
expected that once the natural principles of problem solving were
deciphered, it would bring great improvements to various computational
methods.
Methods of learning, based upon artifi cial intelligence concepts, can be
divided into two major groups: supervised and unsupervised learning. In
the case of supervised learning, there is an input-output relationship
among data that the learner is supposed to learn and afterwards be
capable of correct output predictions for newly introduced inputs.
However, unsupervised learning is characterized by the fact that data
are analyzed with no a priori assumptions of correlation and with
no specifi c goal to predict 'correct' answers, that is, relationship(s)
between the data are revealed after the learning process. Approaches
to unsupervised learning include various algorithms based upon
dimensionality reduction or clustering of the data.
Of many methods based upon neural computing, artifi cial neural
networks (ANNs) are mostly developed and applied in general science.
Their advantages and shortcomings have been discussed in great detail in
many textbooks (Freeman and Skapura, 1991; Veelenturf, 1995; Kröse
and Van der Smagt, 1996; Gurney and Gurney, 1997; Kasabov, 1998;
Haykin, 1999; Gupta et al., 2003; Dreyfus, 2005; Rabunal and Dorado;
2006; Taylor, 2006)
ANNs have the remarkable information processing features of the
human brain, such as nonlinearity, high parallelism, robustness, fault and
failure tolerance, learning, ability to handle imprecise and fuzzy
information, and their capability to generalize (Basheer and Hajmeer,
2000). An ANN is a learning system based on a computational technique,
which can simulate the neurological processing ability of the human
brain (Achanta et al., 1995). The power of neural computations comes
from connecting neurons in a network (Agatonovic-Kustrin and
Beresford, 2000). They can provide answers to 'what if' questions, by
making correlations between input and output parameters (independent
and dependent variables). Since ANNs are used for generalization of

￿

￿ ￿