Fusion Analytics Request for Information (RFI)

Fusion Analytics Request for Information (RFI) April 2018 This is a Request for Information (RFI) synopsis. This RFI is for planning purposes only. This is NOT a request for Quotations or Proposals. No solicitation document exists and formal solicitation may or may not be issued by the Government as a result of the responses to this RFI. The Government will not pay for any response expenses. Interested parties are responsible for adequately marking proprietary or competition sensitive information contained in their response. The Government encourages teaming/partnering. Background The US Army has multi-sensor platforms that are able to capture multiple sources of intelligence data for tactical operations using airborne platforms, such as the Enhanced Medium Altitude Reconnaissance and Surveillance System (EMARSS), and vehicles, such as Vigilant Pursuit. In addition, exploitation of open-source information (both publically available and derived from exploitation of captured devices) will provide additional intelligence data. Improved fusion and pattern analysis of these data are critical elements in order to decrease the user cognitive load by improving fusion algorithms and system ease of use. Fusion analytics introduces capabilities that will provide enhanced analytics and services as well as accommodate data from all the intelligence domains (e.g., HUMINT, GEOINT, and All Source). This RFI is intended to examine the degree to which the fusion solution will be capable of automating the performance of: normalization for all data, correlation for single or for all intelligence disciplines, relationship detection and aggregation, and pattern discovery/exploitation. Description U.S. Army Communications-Electronic Research, Development and Engineering Center (CERDEC), Intelligence and Information Warfare Directorate (I2WD) is conducting this RFI to determine potential capability and innovative approaches for fusion analytics. Data Construct Imbedded in such fusion capabilities is an underlying data construct, for which the government will provide a notional API for information. A central hub supporting fusion capabilities is an entity-relationship datastore that provides information about entities of relevance to the operational goal of situational understanding. To that end, these entities will be of specific types as required by operational need: military units, facilities, pieces of equipment, individual persons, significant events, organizations, personas, and networks. In addition to providing attributes about the entities themselves, the datastore (and API) also needs to provide for interrelationships between the entities, and support attributes on those relationships as well. This data structure can be viewed as a graph, with the entities seen as nodes and the relationships as edges. The nature of intelligence analysis for situation development leads to maintenance of three "layers" of data: raw inputs (sensor reports, text files, video, etc.), entities and relationships extracted from the raw inputs but not yet de-duplicated/correlated, and a - fused' set of entities and relationships that represents a single set of correlated data on each. The attributes maintained in the fused layer will contain data values intelligently combined from the constituent uncorrelated pieces, and specific approaches for this combination can be provided. Provenance ("entity history") is also required between these three layers - raw inputs are connected to their extractions, and the extracted-but-uncorrelated entities and relationships are connected to their fused resultants. Objective Describe new, existing, and/or enhancements of capabilities to achieve the following objectives: Fusion Analytics Capability Areas 1. Extraction of entities and relationships (with type and attributes) from unstructured text Given the types of relevant entities described above, extract from unstructured free text any available entities, determine at some level their entity type, extract available attributes about each entity (including identity, time, location), and identify potential relationships between the entities (to include some description of the type of the relationship). 2. Correlation of extracted entities and relationships (six cases) Correlation is defined as the decision that two or more uncorrelated entities represent the same real-world item. In general, there are six methods which are used today to make a correlation decision: a. Matching of external systems' database keys b. Matching of specific entity unique identity attributes c. Similarity of non-unique entity attributes d. General similarity of entity type/location/time of observation e. Tracking movement of units or equipment observations across time f. Similarity of entities based on common relationships within the graph Where applicable, parameters and thresholds for the correlation calculations should be user-modifiable. After a correlation decision is made, a combination process is required to create the fused representation of the correlated entities (and relationships), and to maintain a history of the connections from raw to extracted to fused data. Because the extracted, uncorrelated data is maintained intact, the ability to manually undo and redo combinations of them into fused records should be supported. 3. Entity graph refinement Assess and refine the fused datastore based on (1) identifying/discovering implied or inferred relationships between the entities, (2) identifying or inferring aggregations or clusters of entities into "larger" entities (like platforms, organizations, higher echelon units, etc.), (3) improving the identity of an entity by detailed analysis (including geo-temporal, multi-source, semantic disambiguation, and other analyses), or (4) identifying significant activities implied by connected events. In all cases, persist the results of any refinement. 4. Pattern recognition and validation (composition, disposition, activity) There are three primary entity patterns of interest: composition (what makes up an entity, such as a unit's owned equipment and subordinate units), disposition (how a group of entities are either physically or notionally arrayed/deployed), and activity (what smaller events can or have aggregated into a larger activity). For these types of patterns, recognize patterns in the existing fused data, and support user validation of those patterns as pattern models. 5. Pattern detection and analysis For pattern models either provided by earlier analysis or derived from #4 above, provide the capability to match (detect) the pattern models in the existing fused data and alert users to the data matching the model. User configuration to specify the level of matching desired should also be supported. 6. Course of Action (COA) development and assessment Courses of Action are patterns of potential activity/movement by forces (with their patterns of composition and disposition) to accomplish their given mission(s). Support generation/development of possible COAs (in term of patterns) and assessment of alternative COAs against the existing fused data. 7. General Data Analysis Support general analytics on the datastore, such as (1) generation of graph metrics (centrality(s), density, connectedness, closeness, transitivity, and others) on selected subsets of the entities/relationship graph; (2) change detection over a selected subset of entities and relationships; (3) specialized analyses over aspects of the fused data (such as mobility/tracking, constructs or activity of communications/networks, geographic or cyber activity, semantic/sentiment analysis, battle damage aggregation, etc.). 8. Pedigree development and maintenance Pedigree of data (either on a particular attribute, or on a collection of them) is defined in terms of accuracy (how well do we know a particular piece of data?) and credibility (how well do we believe a particular piece of data?). Based upon the accuracy and credibility of the information sources, derive an initial pedigree for extracted entity and relationship data. Based upon the pedigrees of the extracted components, derive the fused attributes and maintain the fused pedigree value for a correlation/combination action. Response We are asking industry to provide a response to this RFI that includes: - Provide a Rough Order of Magnitude (ROM) cost estimate for the effort, which includes a 1-2 month effort (e.g., installation, integration, testing, and demonstration). Please include any applicable license fees to test/demonstrate the solution. The government will provide relevant data for testing/demonstration purposes. - Propose a commercial software licensing approach that minimizes license management costs and leverages GOTS and COTS products availability under DoD enterprise license vehicles. - Articulate how your solution applies to the tactical enterprise at various domains and echelons. - Describe the risk areas from a cost, schedule and technical approach with your rationale. Responses to this RFI are to be unclassified and received no later than 01 June 2018. All interested parties shall respond in as clear manner as possible to each of the areas listed above. Information is preferred in soft-copy form in Microsoft Word and shall not exceed 10 pages. You may forward your responses via email address to: usarmy.apg.rdecom-cerdec.mbx.i2wd-fusion-analytics-rfi@mail.mil ; Subject Line: Fusion Analytics RFI. All information must be in writing or via email; telephone requests for additional information will not be honored. You may forward responses to US Army, RDECOM, CERDEC, I2WD, 6605 Surveillance Loop (Bldg. 6003) Aberdeen Proving Grounds, MD 21005, ATTN: RDER-IW-IE (Fusion Analytics RFI). Acknowledgement of receipt will be issued. If you choose to submit proprietary information, mark it accordingly.