| dc.description.abstract | This research is motivated by the need to support inference in intelligent decision
support systems offered by multi-agent, distributed intelligent systems involving
uncertainty. Probabilistic reasoning with graphical models, known as Bayesian
networks (BN) or belief networks, has become an active field of research and practice
in artificial intelligence, operations research, and statistics in the last two decades.
At present, a BN is used primarily as a stand-alone system. In case of a large
problem scope, the large network slows down inference process and is difficult to
review or revise. When the problem itself is distributed, domain knowledge and
evidence has to be centralized and unified before a single BN can be created for the
problem.
Alternatively, separate BNs describing related subdomains or different aspects
of the same domain may be created, but it is difficult to combine them for problem
solving, even if the interdependency relations are available. This issue has been
investigated in several works, including most notably Multiply Sectioned BNs (MSBNs)
by Xiang [Xiang93]. MSBNs provide a highly modular and efficient framework
for uncertain reasoning in multi-agent distributed systems.
Inspired by the success of BNs under the centralized and single-agent paradigm,
a MSBN representation formalism under the distributed and multi-agent paradigm
has been developed. This framework allows the distributed representation of uncertain
knowledge on a large and complex environment to be embedded in multiple
cooperative agents and effective, exact, and distributed probabilistic inference.
What a Bayesian network is, how inference can be done in a Bayesian network
under the single-agent paradigm, how multiple agents’ diverse knowledge on
a complex environment can be structured as a set of coherent probabilistic graphical
models, how these models can be transformed into graphical structures that
support message passing, and how message passing can be performed to accomplish
tasks in model compilation and distributed inference are covered in details in this
thesis. | |
