Data-Centric AI in Multi Domain Awareness

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws. OBJECTIVE: The objectives of this topic are a) to develop, test, and demonstrate a data-centric AI solution for processing Multi Domain Awareness data and b) identify and illustrate the potential negative effects of insufficient training data in automated processing of multi-domain data streams. DESCRIPTION: Enhanced situational awareness and flight safety support in the space domain within and beyond geosynchronous orbit (GEO) is achievable given sufficient data strategies. Observational evidence of spacecraft anomalies is manifested in multiple data types to include raw EO/IR imagery features, astrometric and photometric features, and features associated with their radio-frequency/RF payload. When this data is collected in great volume, from multiple modalities, and geographic locations, there is significant opportunity to enable reliable automated alerting. Today defining normal operations in space so that abnormal behavior can be flagged and more deeply observed is an area of research which addresses fundamental developments required to begin to define a baseline for real-time automated alerting for increasingly crowded orbit neighborhoods, as well as inform how this baseline may be extended to support missions in orbits beyond GEO. This is a significant need as new activities in cislunar space are planned for the coming years. This topic seeks to: Define data-driven methods to drive the development and testing of an ontology of automated alerts in support of government and commercial applications and Develop foundational mathematical solutions needed to enable these mappings on multiple timescales and to properly quantify the uncertainty in these mappings to effectively support decision making. This includes mapping high frequency and geometrically diverse collection data to: Specify/classify maneuver alerts as station-keeping or not, Specify observed anomaly types and classify them on the basis of astrometric, photometric and RF features observed, and c.) Quantify confidence/uncertainty in the mapping from input data to selected alerts. To develop an explainable alert ontology, the awardee(s) will develop a supervised learning approach which combines feedback from experts in space domain awareness with Generative Adversarial Networks (GANs) and Convolutional Neural Networks (CNNs). The awardee(s) will show the regrets associated with insufficient training of SDA models and compare the assessed results of their solution at different levels and qualities of training using available SDA data. Importantly the awardee(s) will identify various forms of data cascades which can occur when insufficient data work is performed in the development of automated processing routines applied to the interpretation of SDA data. When do leaks and false alarms manifest into undesired downstream effects? PHASE I: Awardee(s) will conceive of, develop, and demonstrate multiple methods for automated processing of SDA data which result in meaningful conclusions regarding the observed operations of spacecraft. Awardee(s) will exercise these algorithms using available SDA data. GFE will not be provided. PHASE II: Awardee(s) will extend these methods to fuse multiple modalities and evaluate when and where errant conclusions can be made and what the underlying data conditions are that lead to these non-ideal conclusions. PHASE III DUAL USE APPLICATIONS: Awardee(s) will develop a strategy to transition prototype residual capabilities and incremental proliferation based on operational requirements. REFERENCES: Press, G. (2021, June 16). Andrew Ng launches a campaign for data-centric AI. Forbes. https://www.forbes.com/sites/gilpress/2021/06/16/andrew-ng-launches-a-campaign-for-data-centric-ai/?sh=2f32fc6674f5 Radecic, D. (2022, March 23). Data-centric vs. model-centric AI? The answer is clear. Medium. https://towardsdatascience.com/data-centric-vs-model-centric-ai-the-answer-is-clear-4b607c58af67 Sambasivan N., Kapania S. and Highfill H., "Everyone wants to do the model work not the data work: Data cascades in high-stakes AI", Proc. CHI Conf. Hum. Factors Comput. Syst., pp. 1-15, 2021.; KEYWORDS: machine learning; artificial intelligence; data fusion; space domain awareness;