Machine Learning
The ultimate Information Advantage
The use of Artificial Intelligence (AI), Deep Learning (DL), and Machine Learning (ML) technologies have exploded in the last decade. It has become an integral part of commercial products and services that we use daily, as well as in Department of Defense (DoD) systems.
The DoD has moved quickly to implement AI/DL/ML to gain a competitive advantage. The scope and range of the problems we can solve with AI are unlimited. Agencies can significantly improve operations in virtually every situation, ranging from cybersecurity to combat systems. As the ultimate Information Advantage, Machine Learning can reduce the cognitive load of our armed forces and help them make decisions more quickly and accurately.
A system that learns to improve performance!
DL/ML models do something that other information systems can’t—they learn from operational data. As new data is fed into these models, they learn and provide results that can be used to optimize operations and further improve performance. Some exciting applications that Rite-Solutions is working on with the Navy include:
Ship Classification
Identifying and classifying ships is a persistent challenge for sailors. Using ML models trained with ship images, Rite-Solutions demonstrated that friendly and hostile vessels could be quickly identified and classified with excellent accuracy. For an overview, please visit the Machine Learning Ship-Recognition System page.
Undersea Object Detection
Rite-Solutions is working with the University of Rhode Island via a State of Rhode Island Commerce Corporation Innovation Grant to develop a cutting-edge research approach using new applied mathematical methods and DL to identify objects on the seabed better. The technology will address the problem of identifying three-dimensional objects without regard to their orientation or position, which has clear applications to the undersea detection of mines, shipwrecks, or other navigational hazards or objects of interest.
Automating Data Curation
One of the more significant hurdles for supervised machine learning is the availability of large, labeled data sets for model training. Traditional, manual approaches for curating and labeling datasets are time-consuming, expensive, and error-prone. We demonstrated that weakly supervised learning could significantly reduce the time and cost to label noncurated datasets of ship images needed to train a ship recognition and classification ML model. The benefit of this approach is that new threats will be quickly identified and provided to fleet operators. Please visit our blog, Automatically Curated Data Improves Training of Machine Learning Models to Produce Better Results, to learn more.
