Information fusion is the process of integration and interpretation of
heterogeneous data with the goal to get new information of a higher
quality. The key for an efficient and successful integration and
analysis of data from heterogeneous sources is an interactive and
iterative policy: relevant subsets of data have to be selected,
cleaned, transformed and integrated into a global schema. Next,
analysis and visualization techniques are applied. If the results are
not sufficient, this process will be repeated with different or
additional data, parameters, and methods.
The InFuse system is designed as a database centered and component
based middleware system to support efficiently Information Fusion tasks.
Virtual data integration and analysis: Efficient
support for data integration and analysis of large datasets is ensured
by a tight coupling of the database system and analysis tools. This is
achieved by using the FraQL multidatabase query engine.
Uniform access to heterogeneous data sources: A
CORBA-based adapter architecture and a global view mechanism provides a
uniform abstraction layer to heterogeneous data sources.
Advanced data integration operations: Examples are resolving meta-data and instance-level conflicts
based on queries with "computed" tables and attributes as well as reconciliation functions.
Data transformation: Data
transformation is supported by special operators (e.g. TRANSPOSE),
extended grouping using user-defined functions as grouping criterion
and user-defined aggregates.
Efficient analysis of large datasets: This includes generation of samples from the result of a query as well as selecting the "top k" results.
Multiple views: Various
tabular and graphical views on existing and generated information sets
are available. These support information investigation and exploration
as well as controlling fusion execution.
The global architecture of the InFuse system consists of three main
parts: the fusion engine, the FraQL query processor and a front end for
interactive graphical data analysis and exploration.
The fusion engine represents the central part of the system and is
responsible for different tasks. As a fusion process consists of
several dependent steps, the fusion engine manages the definition and
persistence of processes and controls their execution. Process
definitions as well as the states of running processes are stored in
the central meta data repository. Therefore, several process instances
with different parameter values can be executed for each definition at
the same time.
Special operators (data mining or machine learning algorithms) can be
loaded and unloaded dynamically into the system by the operator
manager. These operators can be thought as a kind of stored procedures
or functions and are restricted to use the mechanisms given by the
Besides these main features, the fusion engine provides additional
services as user and error management. A CORBA based API is provided to
connect to different front ends.
Furthermore, the fusion engine enables the user to perform the
operations on sample data to evaluate for example convenient parameter
settings before executing the analysis for the whole data set. The
intermediate results can be shown and analyzed using various
visualization techniques in the front end. In addition, the results can
be materialized for later use. These features shall help the user
to interactively and iteratively but also efficiently discover new
To support these data analysis techniques on heterogeneous data sources
the fusion engine relies on the features of the query processor called
FraQL. A global object-relational view to heterogeneous date is offered
by the query processor's adapter architecture. Several extensions
in the query language provide easy to use integration and data
Besides the integration steps, efficient data analysis is enabled by
user-defined functions and aggregates, query optimization and
techniques like multidimensional histograms and efficient
sampling. For fast execution times a local cache for condensed
statistics about the data and intermediate results is managed by the
query engine. Hereafter, a virtual integration and data analysis
is made possible by using techniques of the FraQL query processor.
The third part of the project is a front end, which provides a
comprehensive collection of several ways of interacting with the system
and exploring the data. The front end is extensible by a plug-in
concept which allows adding new kinds of views. The tool is a
full-featured administration and execution workbench. As it is
connected to the fusion engine via a CORBA-interface, it can be
replaced by specialized front-ends for different user groups.
- Oliver Dunemann and Ingolf Geist and Roland Jesse and Gunter Saake and Kai-Uwe Sattler.Informationsfusion auf heterogenen Datenbeständen, Informatik,
Forschung und Entwicklung, 2002, Volume 17, Number 3, pages 112-122,
- Oliver Dunemann and Ingolf Geist and Roland Jesse and Kai-Uwe Sattler and Andreas Stephanik.
A Database-Supported Workbench for Information Fusion: InFuse,
Advances in Database Technology - EDBT 2002, 8th International
Conference on Extending Database Technology, Prague, Czech Republic,
March 25-27, Proceedings, Editor: Christian S. Jensen and Keith G.
Jeffery and Jaroslav Pokorn\'y and Simonas Saltenis and Elisa Bertino
and Klemens Böhm and Matthias Jarke, pages 756 - 758, Springer, Lecture
Notes in Computer Science, Volume 2287, 2002, ISBN: 3-540-43324-4
- O. Dunemann and I. Geist and R. Jesse and G. Saake and K. Sattler.
InFuse - Eine datenbankbasierte Plattform für die Informationsfusion,
pages 9-25, Datenbanksysteme in Büro, Technik und Wissenschaft
2001), 2001, Editor: A. Heuer and F. Leymann and D.