Information Technology Reference
In-Depth Information
towards these features can be made, this leads to an increase in the overall model
complexity and/or computational complexity. Temporal orders in input information are
not supported for reasoning by the approach itself. However, temporal concepts can be
explicitly introduced to the system's rule base, which will increase the overall
complexity of the rule base.
The complexity of the model affects the computational complexity for reasoning
only to a small extent, but as rule bases cannot be checked for consistency in an
automated way, the definition and extension of complex models is difficult and error-
prone. Furthermore, rule-based reasoning is well suited to online analysis and is also
scalable to handle large amounts of data.
Case-based reasoning is able to handle incomplete data as well as uncertain data as
input for classification. Temporal ordering of input information cannot be considered in
the reasoning. The scalability of the model towards handling its complexity and the
computational effort for reasoning is neutral, as complex situation models will increase
the computational effort for matching and classifying the current case, but efficient
algorithms already exist for this task.
The expressiveness of models is high, as no generalization in the description of the
classified cases is made and each new case is evaluated in respect to previously
acquired cases. In general, case-based reasoning is suitable for carrying out online
analysis, as efficient algorithms are already available for this task.
Description-logic (DL)-based reasoning is suitable for reasoning of incomplete
information and also extension towards reasoning with uncertainties and temporal
reasoning exits. The expressiveness of the DL-based models is very high. Moreover,
the consistency of the models can be checked automatically, which supports the
definition of very complex models. As complex models seriously affect the
computational effort for classification, a trade-off between the model expressiveness
and the computational complexity of the reasoning on these models must be made. DL-
reasoning is suboptimal for real-time analysis when handling large amounts of data.
The probabilistic reasoning approach of Bayesian Networks (BN) offers a
mathematically founded treatment of uncertain information and can also handle
incomplete information. In the classical approach, no explicit support for temporal
reasoning is given, while there are existing extensions to this. In comparison to other
approaches, Bayesian Networks scale quite well as an inference in a BN is “only” NP-
complete. The expressiveness of Bayesian Networks is probably unsuitable for
representing complex situations or even human behaviour. For online analysis, efficient
reasoning algorithms already exist.
Markov Models allow - like Bayesian Networks - support reasoning of uncertain and
incomplete information. Besides that, temporal reasoning is supported inherently by
this approach. As Markov Models are even more restricted in their expressiveness than
Bayesian Networks, they will not be suitable for recognizing complex scenarios. For
online analysis, this approach is suitable, as very efficient and easy-to-implement
algorithms already exist
Time maps are inherently suitable for reasoning temporal orders of information, but
they neither support the handling of uncertain nor incomplete information. The
expressiveness of time maps and their scalability is limited, and so they are most
suitable for the detection of well-describable and simple situations or activities that
follow a defined schedule. The computational complexity increases non-linearly with
the amount of input information, but online analysis with time maps is possible.
Search WWH ::




Custom Search