Background

Background and Program Perspectives
This ARO research grant is addressing some of the most modern and challenging problems in information processing that face the US Army in its current worldwide operations. It is addressing the overarching issue of automatically or semi-automatically forming the best possible estimates of situational state estimates by Information Fusion operations on a plethora of disparate and uncertain observational and contextual data and information sources streaming in from a dynamically changing operational environment. The complexity of forming such estimates is compounded by the combination of data that is uncertain, ambiguous, and of mixed reliability coupled with the operational problem environment that involves insurgency within a foreign population. Insurgencies and the methods of Counter-Insurgency (“COIN”) operations are extraordinarily complex environments to deal with and even to define. From Army FM3-24 on Counterinsurgency, we have a definition of insurgency as: “Joint doctrine defines an insurgency as an organized movement aimed at the overthrow of a constituted government through the use of subversion and armed conflict”. From the same source, we have the definition “Counterinsurgency is military, paramilitary, political, economic, psychological, and civic actions taken by a government to defeat insurgency” Thus, these conflicts do not involve known, uniformed adversaries, and have very high collateral damage considerations since the conflicts occur within neutral populations. Course of action choices are both highly varied, involving all the factors just mentioned but at the same time is highly constrained. the MURI Problem Domain is considered to be the problem of Small-scale COIN insurgency. In Small-scale insurgencies, belligerent groups have established some size, are developing tactics techniques and procedures, and are causing hostile and possibly lethal events. These groups however are still quite covert and operate very carefully; their leadership and organizational structures and their insurgency-related goals and objectives are still not well understood. Considering the Small-scale COIN problem, the requirements for Information Fusion (IF) are to estimate the “essential elements of information (EEIs)” for this sub-problem space of COIN, in support of corresponding military or other possible courses of action. The framework for research planning for the MURI has thus been developed around a “requirements relevant” but not “requirements-driven” approach to the prototyping of a Hard-Soft IF process; that is, this research program has no operationally-specific Army requirements specification or specific application paradigm. The positive side of this is that the research will not yield a “point design.” For some specific operational application, and should yield an architecture that is flexible to new data sources. However, there is in fact some risk of non-applicability. To deal with this in part the program includes a task to examine scalability and robustness of developed solution strategies. It is intended that these planning aspects be worked in conjunction with the ARO, moving forward. Another critical research strategy choice is that, based on extensive analysis, we have chosen an inductive, learning/discovery-based approach regarding the development of insight for a dynamic COIN problem. Modern literature shows that the ability to effectively model human group dynamics and relationships remains a very challenging problem and that only very limited capability exists. In particular for the Soft Data Fusion problem, the UB team has chosen graph-based methods as an inferencing framework, wherein the soft data are associated and batched into an evolving, accumulating “Data Graph” representing cumulating situational evidence, and using this Data Graph and analyst formed queries that can also be represented as graphs (“Target Graphs”), state of the art methods developed at CMIF are used for Graph Matching to yield inferred assertions, supporting an adaptive analyst learning process. The operational focus here is on human social network type inquiries. On the Hard Data Fusion side, PSU and TSU are combining to use multispectral Hard Data of various modes (e.g. imagery, acoustic, video) to focus on the issue of human-vehicle behaviors and relationships, since vehicles and their use in various ways have proven critical in COIN type operational problems. Using the Uncertainty knowledge base developed by UB in conjunction with IONA college provides a framework for uncertainty alignment in the graph based soft fusion process. This uncertainty knowledge is also part of the hard-soft fusion framework developed by UB. In hard-soft fusion the kinematic tracks with from hard data (PSU) along with acoustic signatures (TSU) are merged with location information for various entities (persons, locations etc.).