Wallaroo Labs on Tuesday announced that the company has been selected by SPACEWERX, the innovation arm of the U.S. Space Force, to solve edge model deployment challenges specific to on-orbit servicing, assembly and manufacturing missions.
Wallaroo’s AI and machine learning platform is designed to accelerate the last mile of machine learning implementation, which is the deployment stage.
“If you think about the life cycle of a machine learning model, first you have all your data wrangling and engineering to get it ready for analysis,” explained Vid Jain, CEO and founder of Wallaroo. “And then you analyze the data to find patterns and build a model that makes predictions based off that training data.”
Those first two steps can be thought of as the first mile, Jain said. Once businesses have this model, they need to think about how to deploy it and get value from it. The last mile is taking a model built by data scientists and then deploying it into production conditions. Then the model is monitored on an ongoing basis to make sure it is still accurate as the environment—and data—changes, he said.
This fully-funded phase 1 project in collaboration with Catalyst Campus (CCTI) will look at edge model deployment challenges for use cases such as satellite life extension, on-orbit refueling, active debris removal and the reuse and recycling of materials to build the foundation for assembly and manufacturing in space, according to Wallaroo.
SEE: Artificial Intelligence Ethics Policy (TechRepublic Premium)
Compute power constraints at the edge
In terms of edge-specific challenges, “the development environment for a model normally involves a data scientist on their laptop occasionally spinning up large amounts of compute power to analyze a batch of historical, cleansed data in order to create a predictive model,’’ Jain said. “But when you deploy it at the edge, the edge has hard constraints in terms of compute power. So it could be a drone or a battleship or a satellite where you have perhaps a streaming video coming in.”
You need a model that can analyze this streaming video and make predictions but there isn’t enough cloud compute power to run the model, he said. “This is where our hyper-efficient, purpose-built engine for machine learning comes in. It allows organizations to generate more inferences on 80% less compute, so they are able to run even complex computer vision or natural language processing models at the edge where compute is restricted.”
Other edge model deployment challenges that Wallaroo helps address include managing model versioning across a fleet of hundreds or thousands, experimentation and testing, model performance observability and deploying to edge locations with inconsistent or no internet connectivity, he said.
Dr. Joel Mozer, director of science, technology and research at SPACEWERX, said the Wallaroo platform was chosen for its modern, interoperable and integrated architecture.
“The mission of the United States Space Force (USSF) is to organize, train, and equip guardians to conduct global space operations that enhance the way our joint and coalition forces fight, while also offering decision-makers military options to achieve national objectives,” said Mozero, in a statement. “To do this effectively, we must invest in AI and ML capabilities that can be deployed in the cloud and at the edge.”
In addition to their work with the public sector, including with the U.S. Air Force, Wallaroo is also working with several Fortune 500 companies to help them deploy and manage their machine learning models at scale, generating better performance and observability over their AI/ML initiatives.
SPACEWERX looked at several well-known cloud and SaaS providers, but Wallaroo was ultimately selected for the scale in which the platform can operate and the reliability offered for their mission-critical deployments, Jain said.
Learn more about Wallaroo in this blog post from Microsoft M12, one of Wallaroo’s leading investors.