Rule-based expert systems solve problems based on rules that are coded
in the form of IF a condition exists THEN perform some action. An
action could be to display a message to the user or to check the
validity of another rule. Conditions are usually facts. So,
essentially the rules check to see if a particular fact is present,
and if it is, the expert system takes the associated action.
Rule-based expert systems learn by being provided with new rules.
Case-based expert systems solve problems by analogy rather than by the
strict application of rules. They are particularly useful when few
facts are available or information is incomplete. The expert system
is presented with cases that describe a situation, its solution, the
results achieved by applying the solution, and key characteristics of
the case that can be searched quickly when the system seeks to match a
new situation it is trying to solve to its database of cases.
Case-based expert systems employ similarity metrics to measure how
similar the problem it is trying to solve is to the past cases in its
database. Case-based expert systems also include adaptation modules
which are used to create solutions for the problem it is trying to
solve "by either modifying the solution (structural adaptation) or
creating a new solution using the same process as was used in the
similar past case (derivational adaptation)." Case-based expert
systems learn by having the human-created solution to a problem that
the expert system could not find a matching case for added to its
database.
"The Basics of Expert (Knowledge Based) Systems" JM & Co/AJRA, 1997
http://www.ajrhem.com/EXPERT.pdf
Rule-based expert systems are most effective when expert knowledge can
be encoded as rules. For example, a rule-based expert system can
complete tax forms very effectively because the tax code is an
entirely rule-based system. Given a set of facts about an
individual's financial transactions and characteristics during the
previous year, a rule-based expert system can, given an appropriately
coded set of IF/THEN statements, accurately determine that
individual's tax liability.
Case-based expert systems are applied when expert knowledge is
difficult to encode as rules and actual experiences need to be
evaluated. One example of the usage of case based systems is in
orthodontics. "Traditional rule-based expert systems have some
limitations when applied to orthodontic diagnosis and treatment
planning. These limitations may be avoided by using a case-based
system which is a particular type of expert system that uses a stored
data bank of previously-treated cases to provide the knowledge for
solving new treatment problems."
"Application of a case-based expert system to orthodontic diagnosis
and treatment planning: a review of the literature." By Hammond RM,
Freer TJ. Aust Orthod J. 1996 Oct;14(3):150-3 Pub Med, National
Library of Medicine http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=9528413&dopt=Abstract
The knowledge engineer plays the critical roles of application
developer and system developer, although these may be separate people.
The application developer acquires knowledge from domain experts,
books, and other useful sources of information, assembles it into a
model of either rules or cases, maintains and enhances the model over
time, and verifies that the model results in the correct answers. The
system developer determines how the user will interact with the expert
system, particularly through the design of the user interface, and how
the expert system is integrated to other systems for the purposes of
gathering data and/or providing its answers for further processing.
"TDDB66 Expertsystem ? metodik och verktyg" by Henrik Eriksson,
Linköping University http://www.ida.liu.se/~TDDB66/slides/Lecture1.pdf
An expert system can only be considered to be successful if it is
correct and consistent at least as often as the human experts it will
be replacing or at least does not result in economic losses that are
larger than the savings resulting from its implementation. Therefore,
it is critical to give the expert system many test problems to ensure
that it performs well, to give it the same test problems multiple
times to ensure consistency of the answer, and to track performance
before and after implementation to ensure that the results are at
least as good as they were prior to implementation and that they are
not degrading over time. It is possible that new circumstances could
arise for which the expert system will not perform well in the absence
of new rules or cases.
American Express tested the effectiveness of its expert system
implementation by comparing the losses experience to after the expert
system was used to those experience when customer service
representatives made the decisions. Not only did it greatly decrease
the amount of time needed to come to a decision, but the expert system
reduced losses from 15% to 4%. However, it is not nearly enough to
check the results once shortly after implementation. Ongoing
monitoring needs to occur so that performance degradation can be
detected early and adjustments can be made to the expert system.
Neural networks are electronic models that seek to duplicate the
structure and problem-solving technique of the human brain. Neural
networks differ from the other approaches to expert system
construction because they are able to train themselves to solve
specific problems through learning by example. Problems are solved by
a network comprising a very large number of highly interconnected
processing elements that work together to solve problems. An
additional benefit to neural networks is that they can maintain
partial problem-solving capability even if large parts of the network
become damaged.
Neural networks are especially effective at identifying patterns and
trends that are too complicated to be noticed by human beings or other
computer-based methods. This ability makes them especially good at
solving problems involving sales forecasting, industrial process
control, analyzing databases of consumer information, and stock market
trading. Here are some additional examples: "recognition of speakers
in communications; diagnosis of hepatitis; recovery of
telecommunications from faulty software; interpretation of
multimeaning Chinese words; undersea mine detection; texture analysis;
three-dimensional object recognition; hand-written word recognition;
and facial recognition."
While many these applications are clearly beneficial, some are
controversial. Facial recognition in particular has drawn a great
deal of attention. It could greatly improve security by identifying
known terrorists as they move throughout the world, but it also
creates the risk of total loss of privacy. Given a sufficient number
of cameras and appropriate software, a person's movements could be
tracked anytime they left their home.
The elimination of the need to develop precise rules and their ability
to continue to function to at least some degree even when damaged make
neural networks attractive for many classes of problems. However, the
examples used to train a neural network must be carefully selected to
prevent the network from functioning incorrectly. Neural networks
also operate unpredictably and can generate unexpected results that
are difficult to explain, whereas traditional computing approaches are
completely predictable."
NEURAL NETWORKS" by Christos Stergiou and Dimitrios Siganos
http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html
For a detailed technical discussion of how neural networks function,
see "Artificial Neural Networks Technology"
http://www.dacs.dtic.mil/techs/neural/neural_ToC.html and "An
Introduction to Neural Networks" by Prof. Leslie Smith
http://www.cs.stir.ac.uk/~lss/NNIntro/InvSlides.html
Sincerely,
Wonko |
