Neurocomputing: An Introduction - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 14

Nowadays, machine learning uses the theory of statistics and neurocomputing in
building intelligent systems, because the core task is making inference from sample
and mathematical models.
Modern computers have revolutionized our globe. They have utterly transformed
the track of our daily life, the way we do science, the way that agriculture and busi-
ness is continued, the way we entertain ourselves and the way we shield out security.
Machine learning is programming the computers to optimize performance criterion
using example data and artificial intelligence techniques. This requires to code the
given learning rules in a appropriate programming language, devise suitable datasets
and write testing/generalization program which output and analyze the result. It is
better to adapt the system for general circumstances, rather than explicitly writing
a different program for each special circumstance. The responsibility of computer
is twofold. First is iteratively train the system using training data to optimize the
parameters of the computing model for solving optimization problem. A learning
routine is able to adapt to the basic characteristic by monitoring the example dataset.
Second is to store optimized parameters and process the huge amount of data for
output and analysis. The efficiency of computing system in terms of time and space
may be equally important as its predictive accuracy in intelligent system design. An
example is an intelligent user interface that can adapt to the biometrics of its user—
namely, his or her accent, facial features, iris, oculomotion characteristics, and so
forth [ 30 ]. The promising capabilities of high dimensional neurocomputing in the
adaptability, context-sensitive nature, error tolerance, large memory capacity, reduc-
tion of the computational efforts, and real-time capability of information processing
in single and high dimensions suggests an alternative architecture to emulate for
machine learning.
1.3.1 Biometric Application
Biometric applications refers to the identification or recognition of individuals based
on unique physiological or behavioral characteristics or traits. The relevance of bio-
metrics with computer and information science is much broader in terms of capability
of the fast computing machines for huge information processing even better than the
human intelligence. In recent past, several techniques and biometric traits are devel-
oped for human recognition. The basic concept is to use special characteristics of
an individual and identify him based on this special property. One can divide spe-
cial characteristics into two categories, viz physiological (face, iris, finger print,
periocular, palm print, hand geometry) and behavioral (voice, eye-movement, gait)
characteristics. In biometrics, the neurocomputing-based techniques are interestingly
used because of better performance over the conventional statistical techniques. The
key reasons that make neurocomputing competitive for biometric applications may
be summarized as follows: