Information Fusion Process Design Challenges for Modern Applications
--------------------------------------------------------------------

The emerging technology of multisensor data fusion 
has a wide range of applications, both in civilian 
and defense areas. Data fusion is an information 
process that operates on all available observational 
and contextual data drawn from some dynamic real-world 
environment in order to develop a best estimate of those 
world states; said otherwise it is an aid to understanding 
application-space situational conditions.  The techniques 
of multisensor data fusion draw from a broad range of 
disciplines, including statistical estimation, pattern 
recognition, control theory, and artificial intelligence. 
With the rapid evolution of computers and the maturation 
of data fusion technology, the door to using data fusion 
in everyday applications is now wide open and presenting 
great opportunities.

Among others issues, this discipline addresses theoretical 
and practical methods by which data is combined from 
diverse sources (sensors and instrumentation or other 
information systems, web sources, as well as from human 
observers) to improve the probability of accurate detection, 
classification, identification, and tracking of individual 
entities of interest, and for understanding inter-entity 
relationships.  System-level design of a data fusion 
process involves source characterization, alignment 
or common referencing of data streams, association of 
the data streams, and data fusion technologies including 
detection (or decision) theory, estimation theory, digital 
signal  processing, and parametric and non-parametric 
data fusion techniques (including computational intelligence  
paradigms such as fuzzy logic, evolutionary computation 
and neural networks) . Other data fusion related areas 
include employment of adaptive control theory for runtime 
sensor management, methods to estimate situation assessment 
(such as danger, conflicts, etc.), and formal statistical 
methods for evaluation of system performance.

The so-called "hard and soft" fusion problem, driven by 
modern applications involving human-based input, as well 
as a need to incorporate web-based data sources, coupled  
with ever-increasing networked-system capabilities that 
make such disparate data are mutually available, has generated 
a potpourri of new challenges for the  information fusion 
community--the "FUSION" conferences have already had special 
sessions on these topics  but solutions are far from being 
in hand, and R&D in these areas continues.

Finally, there is the goal of achieving fully closed-loop 
data fusion process designs, in which conditions in an 
application-space are initially observed and estimated 
by a data fusion process, the estimates used in analysis 
and decision-making processes to select a responsive 
course of action, optimal resources chosen to enable the 
desired courses of action, and finally adjustments to the 
application-space state are (ideally) realized—the fit 
of the data fusion process into this larger closed-loop 
paradigm has yet to be seriously examined.

The talk will be composed of three parts: 
- an introduction to data fusion and current design challenges, 
- hard/soft, and contextual fusion and further design challenges, and 
- a closed-loop process model for situation management and decision support