DARPA launched the program Modeling Adversarial Activity (MAA) program with a goal to develop mathematical and computational techniques for modeling adversarial activity for the purpose of producing high-confidence indications and warnings of efforts to acquire, fabricate, proliferate, and/or deploy weapons of mass terror (WMTs).
MAA assumes that an adversary’s WMT activities will result in observable transactions. While the probability that any one source alone will reveal a WMT threat may be low, the probability of detecting a WMT threat can be increased by appropriately integrating multiple sources of transaction data. The MAA is focused upon the development of mathematical and computational methods that integrate multiple data sources to detect relevant activities and events with high probability of detection and low rates of false alarms.
The US Air Force Research Laboratory (AFRL) has awarded a new contract to advanced software engineering company IvySys Technologies for the Defense Advanced Research Projects Agency (DARPA) programme. Valued at $4.6m, the contract will enable the US software company to support DARPA’s Information Innovation Office (I2O) four-year programme, Modelling Adversarial Activity (MAA).
IvySys Technologies president and CEO Dr James A DeBardelaben said: “It would be difficult to overstate the lack of meaningful test data for evaluating tools that produce indications and warnings of adversarial efforts.
As part of the project, IvySys Technologies intends to provide an automated synthetic data generation capability that would easily integrate observation data associated with terrorist threat activities of interest within a data-dense and realistic background environment. The company is also expected to deliver synthetic transaction data generation software for WMT activity detection applications and realistic data sets comprising ten billion entities and one trillion transactions.
“As a leader in advanced analytics, modelling and simulation, information sharing, and interoperability, IvySys looks forward to building on our partnership with DARPA to confront this intelligence analysis challenge and support the government’s mission to protect our country and save lives.”
Relying solely on synthetic data, MAA Phase 1 is focused on developing the mathematical and computational methods to enable large-scale graph analytics including graph alignment and merging, sub-graph detection, and sub-graph matching. The methods must operate in noisy, complex, and time-dependent environments.
Upon the successful completion of MAA Phase 1, DARPA plans to release a BAA for MAA Phase 2 focus on continued development and integration of methods into a proof-of-concept prototype system.
MAA requires synthetic transaction data to drive the development of techniques and tools in ways that will avoid the privacy and classification issues that can be associated with real-world data. MAA will develop the means to create synthetic transaction data sets that are both realistic and fully releasable to the scientific community, i.e., data that contains neither personally identifiable information nor restrictions with respect to classification.
MAA may draw on related domains, including human trafficking, smuggling of drugs, antiquities or rare wildlife species, and illegal arms dealing, during the creation of synthetic data sets to meet the need for a large amount and diverse types of synthetic data.
Because transaction data may very naturally be modelled using graphs, mathematical and computational methods to enable large-scale graph analytics including graph alignment and merging, sub-graph detection, and sub-graph matching are of particular interest.
MAA will create synthetic transaction data sets of sufficient realism to drive the development of techniques that can perform on real world data. A number of problems share the underlying structure of the WMT activity detection problem in that (1) adversary activities can be represented as a series of actions that must be completed in sequence for the adversary to achieve success, i.e. as a pathway; and (2) many of the actions in the pathway result in observable transactions and the observations themselves may lead to either structured or unstructured data.
In order to detect WMT pathways, MAA performers may need to create (1) a unified transaction oriented synthetic worldview of entities and actions; and (2) pathway models for WMT activities of interest. When a sequence of actions within the synthetic worldview closely matches a modelled pathway, MAA techniques should generate a lead for follow-up analysis.
As noted previously, there are actions in a WMT pathway that will result in observable transactions. Examples of actions of particular interest include purchasing of pathway-related items, shipment of these items to a common location, etc. Detecting a particular pathway will require that these synthetic sources be integrated in a unified synthetic worldview. Creating such a unified synthetic transaction-oriented worldview of entities and actions will require significant advances in graph analysis methods for graph alignment and merging.
A unified transaction-oriented worldview of entities and actions enables the search for pathways using pathway templates that describe methods for acquiring, fabricating, proliferating, and/or deploying WMTs. A pathway template could be a complete cycle for WMT acquisition or a segment of the acquisition. A pathway template provides a mechanism for determining the relevancy of available data. Finding graph structures within the unified worldview that exhibit sufficiently high similarity to WMT pathway templates will require significant advances in graph analysis methods for sub-graph detection and sub-graph matching.