Hybrid
Systems
Description
Hybrid
systems are computational systems that integrate soft-computing
techniques and methodologies such as fuzzy
logic (FL), neurocomputing (NC), probabilistic reasoning (PR), genetic algorithms
(GA), chaotic systems, belief networks, and learning theory. Such techniques and
methodologies are complementary rather than competitive and thus must be used
in combination and not exclusively. This allows us to take advantage of their
respective component strengths and compensate for each other’s weaknesses.
Hybrid systems
allow the representation and manipulation of different types and forms of data
and knowledge which may come from various sources. Refined system knowledge
are used during reasoning and decision-making processes producing more effective
results. Thus, hybrid systems can handle information more effectively and performs
better than pure conventional systems.
Historical
Perspective
Expert
systems, which are aimed to mimic the decision-making ability of human experts,
basically started in the 70's and 80's. The first
generation expert systems, like MYCIN, exhibited knowledge acquisition capabilities
and a certain level of reasoning under uncertainty. However, such systems have
difficulty in handling various types and forms of data and knowledge used by human
beings which are normally vague, imprecise,
and uncertain.
This prompted the
development of second
generation expert systems which are equipped with machine learning tools
that can automatically acquire knowledge from various sources. Additional techniques
for reasoning under uncertainty were also included in these systems.
Hybrid systems
are a natural way to implement second generation expert systems as they integrate
different approaches in one powerful system. It combines different knowledge
representation methods and their associated algorithms used for reasoning.
Hybrid
System Model
Hybrid systems generally
integrate symbolic and connectionist
approaches to encode and refine knowledge. Conventional symbolic AI systems are
strong in representing and manipulating natural-language-like knowledge and presenting
undestandable results while connectionist models like neural networks are powerful
tools for acquiring knowledge from numerical data.
Architectural approaches
to integrate symbolic and connectionist approaches can be classified into the
unified architecture, the
translational architecture,
and the hybrid architecture.
Hybrid architectures can be classified according to the degree
of integration and the mode
of integration.
