|
Network
of Excellence
|
CARiMan
|
The
Internet Portal for Computer Aided Risk Management
|
Responsible: ICSR
Partners: QUB, UNIMI, PW, UoO, all
partners
Objectives
In signal processing and data mining work, CARiMan will seek fundamental
advances in natural disasters, traffic accidents, health monitoring etc.Tasks
here include: modelling of biotechnological, ecological and chemical processes
based on measured signals and expert information; development of conventional
(deterministic), intelligent (fuzzy and neural) models and models of risk management processes taking into account memory
effects for prediction and process optimization purposes. Development of
fuzzy reasoning methods, fuzzy clustering procedures for recognition and
classification of the processed signals. This will enable us to cover both
exact and uncertain information.
Data Fusion
Data fusion from multiple signal sources, and the
integration of online data streams with databases, are major applications of JPA
SP. Data and information fusion may
yield greater signal quality. Fusion
is of course necessary for multiple criterion and multi-objective decision
making. Integration of 3D and other
signals present significant problems, and the frequent need for real-time
performance adds additional layers of complexity. Integration of observable data
and good quality, stored information in databases is necessary for calibrating
and indeed interpreting the real-time data streams.
Regarding
the combination of heterogeneous data we can distinguish two types of
multi-sensory data fusion, active and
intelligent fusion (sequential sensor operation) and data-oriented fusion (simultaneous sensor operation). The former
uses the results of one sensor to drive the operation of the other source(s).
The second scheme adopts the strategy of combining both functional and
structural information to facilitate the detection/estimation task. In this
scheme we employ the fact that sensory data are not independent (revealing a
redundancy of information) but they also contain complementary information that
can provide robust and reliable solutions if the data from different sensors are
suitably combined or fused. The theory of Evidence provides tools for such a
synergetic consideration of different data acquisition sensors, where evidence
regarding the same pathology from two or more sources is jointly combined to
provide increased assurance about the classification results.
Soft Computing
Knowledge
acquisition and management, which are the basis of the proposed CariMan project
rely on two types of information – one received from the sensors as analogical signals and another called expert risk manager’ knowledge collected
and forwarded through the communications monitored system-risk manager (C1) and
risk manager-risk manager/student (C5) and forwarded
to the KB and DB system by the user interface (C4). While
the first one presumes application of data mining and data processing techniques
the second information type is qualitative and subjective that demands methods
of linguistically determined and processed knowledge. However, in both cases the
available information is not always enough or does not exactly match the real
situation due to the complexity and nonlinear behavior of the considered
dependencies. This fact is a basis for implementation of intelligent (soft
computing) methods for solution of the signal processing, classification and
recognition tasks. Intelligent methods comprise fuzzy set descriptions and
applications, neural network and fuzzy-neural techniques as well as genetic
algorithm implementations. They are powerful tools for system description in
case when the available information is uncertain, biased or subjective.
Reasonable
implementations of intelligent methods in CARiMan refer to description and
recognition of the current monitored system state based on the real time
monitoring and data classification. The developed algorithms and models for this
purpose will provide advanced framework for reliable supervision, classification
and decision making.
Secondly,
the aim of soft computing methods is the development of procedures for
representation of the available expert knowledge. By means of these algorithms
the subjective information received from the experts through C4 will be
implemented in the Database and Knowledge Management System.
One
challenging aim of the advanced monitoring, supervision and decision support
must be the simultaneous use of ‘all’ available knowledge. We cannot afford
ignoring a big part of the available knowledge. According to this view it will
be of a crucial importance to build more then one model (deterministic, soft
computing, stochastic) for recognition and evaluation and then to combine the
predicted values. Alternatively the building of hybrid models that combine
different types of information could be also helpful. That
is way the results obtained by the soft computing methods should be compared
with the traditional methods and thus both results should be flexibly accounted
for in the procedures of decision taking. This
will increase the accuracy of the obtained results.
Data Mining and Knowledge Extraction
In data mining, the CARiMan network will include:
-
knowledge
management,
-
analysis,
design and applications of cryptographic techniques,
-
pattern
recognition and signal processing and classification.
Some of the applications concern the domain of medical
image processing, for example an automatic microscope cell image processing for
immunology analysis. The fields of interest are metaheuristic methods for
solving NP-hard combinatorial and global optimization problems, more
particularly, the variable neighborhood search metaheuristic with applications
in solving clustering and pattern recognition, data mining, location,
transportation, network optimization, etc.
Visualization
Nowadays data visualization and correct image analysis
may deal to effective risk management. The risk manager may be assisted by a
computer based environment helping him in enhancing visualization and correct
data and image interpretation. More in detail full 3D visualization, like
volumetric viewing and computer driven image processing and restoration may
greatly improve the final decision making, avoiding negative situations.