Philosophies and Methodologies for Knowledge Discovery, Deployment, and Development of Decision Support Systems (PMKD4'06)
Krakow, Poland: 4 - 8 September 2006
|
|
|
|
Keywords: metadata, meta-models, semantics, ontology, measurement theory, models, knowledge, meaning, meaningfulness, epistemology, knowledge-discovery, (...creativity, perception, consciousness?), deployment, Decision Support Systems. Many of the issues considered in the main DEXA and DEWAK
conferences relate to sophisticated technical developments relating to
particular problems, algorithms and applications of data mining and knowledge
discovery techniques. In contrast, this Workshop
focuses on the general philosophical and methodological issues underlying the processes of
knowledge discovery, and the emerging technologies for their substantiation. All data mining algorithms are based on families of descriptive or predictive models which implicitly contain within them a view of the “real-world” (i.e. an Ontology). Such views involve concepts of entities in that real world, their attribute sets, the relationships between them, and a measurement theory for how data is obtained. The relationships between attributes, variables, and measurements stored in databases needs to be clearly and meaningfully described and captured in knowledge discovery. The
necessary and logical relationships that exist between attributes of entities in
the real world, and the measured attribute names, and relational structures
embedded into a database of data from the real world embody the Semantics
relating to metadata, another theme of the Workshop.
Explication of such world views of existence and frameworks for
conversing about them, for data mining and knowledge discovery techniques, in relation to metadata and
its semantics, is clearly important and is also one of the primary themes of the
workshop. However, the real-world is
extremely complex, as are the large and complex databases of data that are
collected from them. Models of the
real-world calibrated/trained/fitted to empirical data are encapsulations of
knowledge about the real world. However, such models will necessarily be
simplifications or approximations to relationships and processes of reality,
with the consequence that all data-derived world-view knowledge-model
representations are only ever partial, and hence essentially erroneous to some
extent. Reality also seems to
contain such heterogeneity that we ascribe “randomness/stochasticity”
to many events and processes, and it is important to distinguish the differences
between the deviations of data from the fitted/trained/calibrated model which
may be ascribed to the model lack of fit to reality, and those deviations which
arise from the essential stochasticity of the real world.
This stochasticity itself has a structure in the real world and obtaining
knowledge models of this stochasticity structure is also an important feature of
knowledge discovery. All of these considerations may be put under the general
heading of ‘Theories of Knowledge and Meaning’, or Epistemology,
another main theme of this Workshop. Notation and symbolism are
powerful intellectual aspects of model formulation and representation which can
have an influence on the interpretation of the meaning of
discovered knowledge. This area, Semiotics, is therefore another
topic for this workshop. Knowledge discovery in
large and complex databases cannot be limited to the training of a predetermined
and static family of models/algorithms. Automatic machine-based search, discovery and learning strategies need also
to be modelled. One approach is to attempt to generalize the way in which humans search for and discover knowledge.
Another approach might be to develop new algorithmic methods as has been done by
Google and Cisco for. the internet and www as large and complex
networks.
Artificial Neural Nets, (ANNs), Genetic Algorithms, (GAs),
Learning and Control Theory, Autonomic Systems and
Intelligent Agents all may be regarded as dynamic adaptive heuristics for
knowledge discovery. It is important to consider questions of knowledge and
meaning, discussed above, in relation to such machine-learning technologies. However, we must not loose sight of the fact that the KD enterprise is initiated by the desire to obtain useful knowledge which may be integrated with prior knowledge in a way which will enhance future decision making. Issues of Deployment and Management of Knowledge within appropriate Decision Support Systems are therefore of crucial importance. All the philosophical/technological issues mentioned above relate directly to issues of data mining, knowledge discovery and deployment for decision making. However, it is clear that they are also the issues which underpin the major topics of human and machine perception, consciousness, self-consciousness, and even free-will and creativity. It might be the case that the some contributors to Workshop will find it necessary to consider these issues as they examine the future of knowledge discovery and decision science. Preliminary Themes/Topics 1. Methodologies for
Data-mining and Knowledge Discovery. |
