 |
Piranha
There is a massive amount of intelligence data available that
cannot be manually analyzed. Computers can provide some
help in this problem, but the shear volumes of data make the
most promising approaches impractical. The challenge is for
a computer to sift through a large amount of data & provide a
human with accurate and relevant information, not to merely
allow the analyst to search over an ever increasing set of
data. This requires software that is able to filter, relate, and
show documents & relationships to an analyst.
Piranha Fact
Sheet Piranha
White Paper
|
 |
CIPHER: Counterintelligence Penetration
Hazard Evaluation and Recognition
A great deal of very sensitive information (from
personal credit card information to nuclear weapons
design) resides on a very wide collection of
computer networks. Various illicit groups use a
wide variety of means (unsophisticated, semisophisticated,
and highly sophisticated attacks) to
gain access to this highly sought after sensitive data.
The unsophisticated and semi-sophisticated types of
attacks can be easily identified. However, there is no
method available to prevent the “low and slow”
attacks from sophisticated attackers.
CIPHER Fact Sheet
|
 |
I2IA: Image to Intelligence Archive
An enormous volume of geographical data is being produced
on a daily basis throughout the world and is being analyzed
to create scientific, military, and intelligence information.
Two significant challenges that exist in producing this image
information are: managing the vast amount of available and
increasing data, and automating the manual processes that
are currently needed to produce and search this type of
information.
I2IA Fact Sheet
I2IA White
Paper
|
 |
ORION
Actionable intelligence is critical for threat-vulnerability
analysis and for assessing potential terrorist threat scenarios.
However, the lack of an automated information discovery and analysis
system that allows fast and effective retrieval, analysis, and
fusion of information severely weakens the effectiveness and
efficiency of using all available information for decision-making
and threat analysis. Currently, experienced analysts must perform
fusion of such information. As a result, much of the staggering
collection of information is not utilized or significantly
underutilized. Utilizing ORNL’s expertise in information analysis
and fusion techniques, these challenges can be met using an
agent-based information analysis and fusion system.
ORION Fact
Sheet ORION
White Paper
|
 |
Agent Simulation
Scientists who use simulation models to better understand
physical phenomena commonly deal with massive datasets.
The output of such a simulation can often be terabytes in
size, widely distributed, and may require months of
supercomputing time to produce.
Fact Sheet
|
 |
Machine learning
At the foundation of many applications that perform analysis
over sets of natural language texts lies the task of extracting
information into a structured form. Despite some demonstrable
successes, Information Extraction (IE) suffers from a major flaw
in most real applications. The extraction task for which a tool
was built is rarely identical to the task on which it is deployed,
and shifting IE tools to new textual domains (e.g. from newswire
to emails) results in significant performance drops, even for
simple types of extraction and even for slight shifts in domain.
The errors propagate through multiple subtasks resulting in
even more significant performance reductions for more
complex tasks. Modifying extraction systems to work on new
domains or new tasks has traditionally been a tedious process
and the cost was not always justifiable.
Fact Sheet
|
 |
Ant Swarming
Some problems are too complex to be solved by agents’ work individually.
The emergent behavior (EB) provides a synergistic ability for a collection of
intelligent agents by producing better solutions to a problem than the sum of
the abilities of all agents when they work individually. However, designing an
agent system with EB properties is different than designing the traditional
multi-agent system (MAS). In traditional MAS, the problem solving solutions are
pre-programmed inside each agent. However, in an EB enabled agent system,
neither individual agent has enough intelligence nor any goal or intention to
solve the problem. The problem’s solution emerges from the agents’ direct or
in-direct interaction(s). Scientists and/or engineers are challenged to build such
a system.
Fact Sheet
|
 |
GPU Text Analysis
In the last decade, an explosion in the amount of available digital text
resources has occurred. It is estimated that the Internet contains
hundreds of terabytes of text data, a sizable amount of which is in an
unstructured format. We will soon reach a point where terabyte-scale
text corpora are routinely used on personal desktops for the
purposes of research and decision making. However, most current
text processing algorithms work well only on small corpora and are
difficult to be scaled to the terabyte level on desktops because of the
lack of enough computing power. Even running some simple text
analysis tasks can take days or weeks of computer time to process
a relatively large collection of data.
Fact Sheet
|
 |
Threat Assessment
Assessing the likelihood of a situational threat is challenging
due to the complexity of merging data from disparate sources
and different formats. In addition, those factors in the situational
context that affect the accuracy of observed measurements must
be considered.
Fact Sheet
|
