-
Online Knowledge Discovery from Sensor data for security applications
Detecting anomalies based on sensor readings and other relevant features
is a key challenge in building regional transportation corridors. The
primary requirement from such a detection system is to assist
law-enforcement officers in real-time decision making. Preconceiving a
definition of “anomaly” and building a rule-based system according to
that is naive for most anomaly detection tasks.
-
Multivariate Dependence in Climate Extremes
The objective of this project was to develop novel methodologies to
expand our understanding of extreme values and nonlinear processes from
disparate real-world data and to apply these methodologies to climate
and geophysical data, including real observations from remote and/or in
situ sensors, as well as simulations from state-of-the-art climate or
geophysical models. We investigated the linear or nonlinear dependence
among the usual and the extreme values from time series, spatial,
space-time, and geographic data in the presence of noise resulting from
model or measurement errors, in the context of both large and limited
data sizes. In particular, we focused on hydrologic extremes and the
climate-hydrology connections. Our specific interest was to determine
the impact of climate anomalies on precipitation extremes and river
flows, as well as the spatio-temporal trends of, and relationships
among, precipitation extremes.
-
Monitoring land cover from
remotely sensed data
In geographic applications, change
detection plays a critical role in land use/land cover change analysis
such as monitoring deforestation or vegetation phenology. Coupled with
recent advances in sensor technology, huge amount of land cover
information is now available and accessible. It can be expected that
real-time detection of land cover change will attract great demands in
the near future since it has wide applications in disaster response (for
example, aftermath of nature disasters like hurricanes or earthquakes as
well as possible impacts from man-made disasters like "dirty" bombs),
urban planning, and warfare scenario assessment.
-
Computational Social Science for Threat
Cognizance
The goal is to model a social network
and the behavior of the entities involved in it, given partial
observations. In a real word scenario, observations might be given by
gathered intelligence or news events. We have proposed to approach this
problem using dynamic graphical models.
Figure: Results of Label Propagation on Blogs: (a) True political
affiliations (yellow indicates unknown), (b) Hand-labeled complete data
used for experiments (yellow nodes were assigned labels after reading
their blogs), and (iii) Predicted output of label propagation (red/blue
are the known initial labels, light red/light blue indicate predictions
made by label propagation).
|
|
|
|
|
Point of Contact:
Auroop R. Ganguly,
Geographic Information Science & Technology, Computational Sciences and
Engineering Division, Oak Ridge National Laboratory
1 Bethel Valley Road, MS 6017, Oak Ridge, TN 37831, Phone: +1-865-241-1305,
Fax: +1-865-241-6261, Email:
gangulyar@ornl.gov, Website:
http://www.geocities.com/auroop_ganguly
|
|
