The Office of Naval Research (ONR) Machine Reasoning and Learning program addresses topics important for efficiently obtaining and exploiting information inherent in complex real-world data to gain situational understanding.
The representation of real-world information sources involves the development of automated systems for supporting efficient storage, retrieval, conflation and deflation of heterogeneous data. Such data, whether derived from conventional sensors, intelligence or open sources, are inherently uncertain, incomplete, imprecise and contradictory (UIIC).
Any method for data representation must also represent UIIC effectively. The representations must be linked to the goal of mission-focused autonomy and must be computationally efficient. For every situational awareness problem, representations are also required for assumptions, knowledge, activities and events, as well as information that is typically described qualitatively.
The impact of UIIC on the representation and decisions must be quantified. ONR seeks methods that understand how—for a given mission—data and information should be combined to achieve mission-aware cognition of the environment in the presence of UIIC. The system should be adaptive with a capability to acquire new data to improve situational understanding, with required fidelity or precision linked to the context of the mission and inference task that is being performed in support of the mission. The system also must have the ability to combine and interpret the data faster than the operational tempo so that important activities, events and interpretations of intent and threats are not missed. This is particularly important for situations that are complex and fluid, featuring transitory activity and self healing networks.
The system must be capable of automatically providing multiple hypotheses consistent with the data, commanders intent, and developing mission strategies to resolve UIIC and enable analysis that reveals the intent of objects within the mission context. This implies a need for the system to understand and correctly interpret the intention and information needs of the human components of the system. A desirable property for the system is to detect and recognize the emergence of novel situations and to provide alerts to the human decision-makers.
Integration of the components into a scalable system with predictable properties and guarantees of correctness that support operations over multiple temporal and spatial time constants and interactions with humans are also desirable properties and present significant science and technology opportunities.