Enduring C5ISR Data Hub for Algorithmic Warfare

February 26, 2019 Data & AI, MarkLogic

In our previous post, “Innovation for Activity-Based Intelligence,” we introduced Earthcube—an innovative startup using Artificial Intelligence (AI) and machine learning to revolutionize Activity-Based Intelligence (ABI). While most Algorithmic Warfare capabilities are quarantined to the lab, Earthcube is leveraging MarkLogic® to develop a platform that is able to deploy multiple levels of AI over the same common operating picture to serve a large variety of missions.

Emerging Algorithmic Warfare Challenges

It took decades for military and intelligence systems to evolve from single-purpose hardware platforms to truly net-centric capabilities exposing a unique theater of operations view across all weapon systems. Most will argue that we are still not there and are still operating while fighting ISR silos. Since the US and NATO allies are investing in AI to field new Algorithmic Warfare CONOPS, not much attention is being paid as to how AI agents are going to cooperate across missions, share the same operating picture and be cost-effectively trained, tested and fielded from the cloud to the tactical edge. Beyond operational concerns, the biggest challenge for the commands is going to be how to leverage AI without having to require a Ph.D. to complete basic training.

Earthcube’s Innovative Approach

Earthcube’s approach is guided by a deep understanding of current remote sensing imagery and unmatched AI and machine learning expertise. Their product seeks to empower analysts who are not data sciences experts to autonomously select, train, deploy and test AI agents in support of multiple missions such as site monitoring, change detection and object and activity identification. The AI agents run on the same distributed production system and can collaborate on complex tasking from the same operating picture.

The following pictures show examples of different aircraft class detection.

Earthcube has been able to create one of (if not the) largest database of military objects labeled at the pixel level. All objects and activities are labeled using DoD and NATO order of battle standards, and more than one million of them are stored in a MarkLogic-based C5ISR Data Hub and securely shared across all Processing, Exploitation and Dissemination (PED) workloads.

Earthcube’s Enduring ABI Framework

To empower operational analysts, Earthcube developed a proprietary AI framework and production pipeline to generate datasets and train, test, deploy and maintain AI applications.

Analysts can select AI agents amongst best-of-breed solutions and load a specific dataset into a labeling user interface that allows the analyst to quickly create training sets.

Labeling User Interface

Available label types include vehicles (three classes), aircraft (five classes), helicopters, ships (five classes), tents, buildings and roads. Labels are used to train the AI against a mission’s specific information. The training phase is achieved through a customizable model that covers the three blocks of object detection (segmentation, instance segmentation and object detection).

Object Detection
Instance Segmentation
Segmentation

After this training, the AI agents are deployed to process live data. Observations, labels and objects properties are stored in the C5ISR Data Hub that is built on MarkLogic. It allows Earthcube to quickly add new ISR sources and adapt to ever-changing ontologies. All data and models are natively built on an AWS infrastructure and synchronized in disconnected, intermittent, limited bandwidth (DIL) environments. With this framework, analysts have the ability to field new detectors in a matter of weeks (not years) and create dedicated alerts in relation to activities on monitored sites.

AI as a Service

Earthcube’s platform is available as a service and can be deployed in Amazon AWS to leverage commodity compute and storage to field AI capabilities globally.

Looking Ahead

Earthcube plans to further automate the process of creating and maintaining AI agents to bring the amount of required human intervention and relevant expertise to a minimum. Their work will include creating relevant technologies and processes for automatic labeling as well as active and continuous learning.

For More Information

Idriss Mekrez

Chief Technology Officer, Public Sector