In general we have found that, as expected, information dependency measures
are valuable for characterizing the general "landscape" of
a relation instance.
They are potentially valuable in drill-down situations as well,
where a specific investigation makes the computational cost worthwhile.
More interesting are the indirect insights, such as new approaches
to table decomposition; insights which do not directly depend upon
the measures but are suggested by considering aspects of the measures.
We have also discovered that
information dependency measures are but a special case in a framework
for measures in information systems. This framework allows for the
general study of the properties of measures and how they may be
applied to specific problems in databases and data mining. Of
particular focus is the applicability of the frequent itemsets problem
in data mining.
As a result of this project, three PhD students have finished,
a fourth will finish during the current year,
and others are in the pipeline.
Several MS students have performed experiments relating this topic.
Lecture materials for undergraduate classes have been developed, covering
background and
current research results