Computational Intelligence: An Introduction - Computational Intelligence in Time Series Forecasting - page 7

1.5 Computational Intelligence
According to the published sources, the term computational intelligence was
coined and defined by Bezdek (1992a), in his attempt to study the relationship
between neural networks, pattern recognition, and intelligence. He stated that
computational intelligence deals with the numerical data provided by the sensors
and does not deal with knowledge. This is different from artificial intelligence,
which mainly deals with the non-numerical system knowledge.
Bezdek later attempted to classify the two kinds of intelligence, considering
artificial intelligence as a “mid-level computation in the style of the mind”,
whereas computational intelligence was the “the low-level computation in the style
of the mind”. However, this classification and the definitions of two types of
intelligence, viewed more or less from the aspect of pattern recognition and neural
networks, remained as more of a personal view of the author than a general
opinion.
A still different view on computational intelligence was presented by Poole et
al . (1998), who considered computational intelligence as the study of intelligent
agent design, i.e. capable of learning from experience and flexible to the changing
environments and to the changing goals.
However, a most decisive step in defining the nature of computational
intelligence was made during the 1994 IEEE World Congress of Computational
Intelligence (WCCI), which brought together the International Conferences on
Neural Networks, Fuzzy Systems, and Evolutionary Programming. On the eve of
the WCCI, Marks (1993), in his Editorial to IEEE Transactions of Neural
Networks entitled “Intelligence: Computational Versus Artificial,” pointed out that
“although seeking similar goals, computational intelligence has emerged as a
sovereign field whose research community is virtually distinct from artificial
intelligence”. This indicated that there are two alternative intelligent technologies,
the artificial and computational.
In the middle of the 1990s, some researchers advocated defining computational
intelligence using the adaptivity concept. Eberhard et al . (1995) pleaded for a
definition of computational intelligence as a methodology that exhibits the
capability of learning and that comprises practical adaptation concepts, paradigms,
algorithms, and implementations for facilitation of appropriate actions in complex
and changing environments. Similarly, Fogel (1995) suggested that the intelligent
technologies, i.e. neural, fuzzy, and evolutionary computation, brought together
under the generic term computational intelligence should be viewed as a new
research field holding the computational methodologies capable of adapting
solutions to new problems without relying on human knowledge. Bezdeck went a
step further and even viewed computational intelligence and adaptation as
synonyms.
To sum up, in the last decade or so, we have witnessed a parallel evolution of
two computational streams, soft computing and computational intelligence, both
based on methods and tools of artificial intelligence (Popovic and Bhatkar, 1994),
predominantly on neural networks, fuzzy logic, and evolutionary computation.
Nowadays, because both soft computing and computational intelligence have
integrated a large number of computational methodologies, it is difficult to draw a