Like elsewhere, Artificial Intelligence is dramatically changing the way geospatial intelligence is collected, analyzed and shared. And that’s a good thing – because every day, terabytes of information-rich data are collected from cameras, sensors and satellites.
Artificial Intelligence and Machine Learning (AI/ML) models allow agencies to ingest data faster, mine it more accurately, and identify what needs human attention. This ensures human-in-loop intelligence quickly makes its way to decision makers in situations where every second matters.
Of course, you can’t just point an AI/ML solution at a dataset and wait for the magic to happen. Instead, teams must develop, train, and test algorithmic models that understand the data in front of them and can surface essential information. This kind of human-led data labeling enables a higher degree of model accuracy and aids in achieving fair and unbiased annotations. This affords an additional layer of transparency that supports ethical AI.
Our teams work collaboratively with agency analysts to build and refine AI models for geospatial use cases. We bring together open-source information, research, and commercial tools, along with industry best practices, to create solutions that can be adapted for sensitive intelligence purposes.
We then test those tools and incorporate our findings into our work with customers. In this vein, we act as the connective tissue between customer data and public assets. For our geospatial customers, this type of collaboration enables us to quickly create new geospatial algorithms and models that are more effective and use more data with greater accuracy. We test different algorithmic approaches, adjusting weights and biases, and engineer efficiencies that are faster and less compute intensive.
This is all made possible by our DeepSky Laboratory, which mirrors government environments and allows teams to test new capabilities and collaborate constructively. The lab brings together technology partners, industry partners, emerging technology companies, and academia to incubate and prototype new solutions. The results reduce risk for customers and accelerate the development of new capabilities that advance their missions. Teams from around the world can access DeepSky remotely and collaborate in real-time, accelerating innovation and feedback cycles, and effectively demonstrating a solution before it is implemented in a customer’s environment.
Working together in this way allows us to push the technology forward, continuously make improvements, and stay current and adaptive. In turn, this has led to tremendous customer intimacy and mission success, which allows us to iterate on solutions and apply them to real-world, on-the-ground scenarios.
On AI/ML modeling in particular, the DeepSky lab is an important place to experiment with things like data labeling, a precursor to training the AI/ML models we develop. We have set up tools specifically for testing that use training data for labeling purposes. The purpose is two-fold: It offers us an avenue for experimentation with the latest tools and services that we may use with customers, and it also enables us to train and modify our algorithms. The data labeling process typically begins with an Area of Interest (AOI) and the related geospatial imagery. Objects of interest can then be manually labeled to train AI/ML models. However, manual labeling can be time consuming and labor intensive.
As part of process evolution, a model that has been trained on test data, and has been tagged by AI, can be developed faster and subsequently can be trained on a larger dataset. Experimentation with research, commercial products and approaches – and doing so in a collaborative, lab environment like DeepSky – enables us to develop high quality AI/ML models for the customer. It allows our teams to stay up-to-date on cutting edge academic research in order to diversify our experimentation. It also means our teams can stay informed on the latest theories and mathematics behind successful algorithms. This all allows our models to handle more data, with more accuracy, and with more success – which is the goal we strive to fulfill as a mission partner.
Bringing together expertise, best practices, learnings from across our business, and fresh tools and approaches from the commercial landscape is part of our value to customers. Providing a place for us to jointly experiment and quickly develop new solutions is where we shine. Doing both – and doing it better and better all the time – is truly delivering the art of the possible.
