|
Rapid Identification of Asymmetric Threat Networks from Large Amounts of Unstructured Data
Navy SBIR 2008.1 - Topic N08-084 ONR - Mrs. Tracy Frost - [email protected] Opens: December 10, 2007 - Closes: January 9, 2008 N08-084 TITLE: Rapid Identification of Asymmetric Threat Networks from Large Amounts of Unstructured Data TECHNOLOGY AREAS: Information Systems, Battlespace ACQUISITION PROGRAM: PMA 480 AT/FP(Anti Terrorism/ Force Protection); ACAT II The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), which controls the export and import of defense-related material and services. Offerors must disclose any proposed use of foreign nationals, their country of origin, and what tasks each would accomplish in the statement of work in accordance with section 3.5.b.(7) of the solicitation. OBJECTIVE: Develop algorithms and technologies that will rapidly process massive amounts of unstructured data in order to expose at risk populations and determine the likelihood of threat activity in time to warn or intervene. It is most desirable that the large volumes of data be processed on laptop computers forward deployed as opposed to reach back nodes. The technical objective is to develop neural networks that are able to understand how word, entity and theme occurrence can be used as a predictor of risk. Developed neural network based analysis tools need to be able to model a new domain quickly, requiring only months of past data. Once trained, the analysis tool should be capable of monitoring a large number of processed open sources in real time, monitoring trends and threats. Warnings and intervention shall consist of determining impending threats in order to quickly generate the actionable intelligence needed to accelerate the friendly strategic communications and/or force planning decision-making and execution cycle. This faster cycle will increase friendly operational tempo, depriving asymmetric and irregular foes of the initiative and forcing them into reactive modes, thus rendering them susceptible to disruption, manipulation, and defeat. Databases to be considered are open source intelligence (OSINT) including BLOGs and other available unstructured data sources. DESCRIPTION: Most unstructured inputs are currently analyzed using manual or automatic extraction technologies in order to identify entities of interest and link them in relation to times, places and events. Unstructured data is numerically reduced and inputted to graph analysis systems. This SBIR will explore automatic analysis of reduced unstructured data using artificial neural networks in order to enable trend analysis of at risk populations and the measurement of threat levels. The inputs to the neural net can be themes, specific words or bundles of words, mentions of entities, or entity relationships found in open source references pertaining to an area of interest. Anticipated asymmetric and irregular adversaries are ever-changing entities, often seeking to hide among indigenous populations and exhibiting decentralized, yet self-synchronizing, network structures, as well as the ability to quickly adjust their techniques, tactics, and procedures (TTP) in response to U.S. actions. Future threats will most assuredly be more adaptive, deadly, complicated, and harder to discern than those the US currently faces. It is understood that blind extraction of relevant predictive information (as actionable intelligence) from such a large set of data is an inherently complex computational problem. However, the potential for approximate but fast and robust computational algorithms exists. Neural networks are generally relevant to this class of problem. PHASE I: Develop an approach that will support the processing of large streams of unstructured data using artificial neural networks. The approach should be efficient and timely to minimize processing power requirements to enable real-time or near real-time operation. The offeror may use as is or improve existing open source word, entity and theme tools to create the structured input needed by the analysis tool PHASE II: Develop a prototype system that uses artificial neural networks to predict trends in at risk populations and threat levels by processing large unstructured data streams in near real time. The offeror may use a mixture of existing tools and new tools to rapidly create structure out of large amounts of unstructured text. PHASE III: Demonstrate the products developed under Phases I and II can perform in operational systems via real field or simulated demonstrations. Provide software and hardware packages for field use as appropriate. PRIVATE SECTOR COMMERCIAL POTENTIAL/DUAL-USE APPLICATIONS: Homeland Security initiatives are driving municipal, county, state, and federal agencies to obtain an interoperable communications capability. The technology developed from this topic is directly applicable to these non-DOD threat-warning applications. REFERENCES: herve/abdi.primer.pdf 2. Bar-Yam, Yaneer (2003). Dynamics of Complex Systems, Chapter 2. http://necsi.org/publications/dcs/Bar-YamChap2.pdf 3. Bar-Yam, Yaneer (2003). Dynamics of Complex Systems, Chapter 3. http://necsi.org/publications/dcs/Bar-YamChap3.pdf 4. Bar-Yam, Yaneer (2005). Making Things Work. http://necsi.org/publications/mtw/ 5. A close view to Artificial Neural Networks Algorithms (2007). http://www.learnartificialneuralnetworks.com/ KEYWORDS: fast, robust, artificial neural network algorithms TPOC: Martin Kruger
|