Maxar’s data analytics division adds jobs to sate U.S. government appetite for artificial intelligence

by
Geospatial data for AI training is a booming market. Radiant Solutions president Tony Frazier: “This is an area of intense focus.”

WASHINGTON — Radiant Solutions plans to add 300 data scientists, software developers and geospatial analysts to its workforce of 1,100 over the next year to meet a growing demand for military intelligence and mapping.

A division of Maxar Technologies, the Herndon, Virginia-based government contractor, like others in this sector, is benefiting from heightened interest in artificial intelligence both in the public and private sectors.

“We are seeing strong growth across the intelligence community and the Department of Defense,” Radiant’s president Tony Frazier told SpaceNews.

Of the company’s $300 million in annual revenue, 90 percent comes from U.S. government contracts with DoD, the intelligence community and U.S. Special Operations Command.

One burgeoning market associated with the rise of AI is training data. “Our goal is to make data openly available to facilitate the creation of great algorithms that we can then apply at scale against commercial and government sources,” Frazier said.

The Pentagon got its first major taste of AI from Project Maven, a year-old venture with Google. DoD awarded Google a contract said to be worth about $10 million to help develop machine learning algorithms to analyze live video streams from Air Force drones and identify targets. Battlefield commanders got unprecedented real-time intelligence that typically would take geospatial data analysts days or weeks to produce. The project has been fruitful for DoD although controversial. The New York Times reported that Google’s partnership with DoD will end following backlash from company workers.

Pentagon officials have been explicit about their plans to weaponize AI. Defense Secretary Jim Mattis told lawmakers DoD will consolidate AI projects into a central program office. Undersecretary of Defense for Research and Engineering Michael Griffin has put AI on the list of top technology priorities for future investment.

Frazier said DoD and the intelligence community are seeking technologies to automate data collection, extract information from data from both government and commercial sources. Competitions known as “machine learning challenges” — sponsored by the CIA’s investment arms In-Q-Tel and IARPA, by the Pentagon’s Silicon Valley office DIUx and by the National Geospatial Intelligence Agency — are “indicative of the quest to transform the mapping and military intelligence mission,” Frazier said.

A project called SpaceNet was funded by Radiant and other companies to help automate mapping of roads and buildings. IARPA made data available to automate discovery of mapping features. DIUx and NGA sponsored a challenge to automatically count 54 types of vehicles to help expedite relief efforts on the ground after a natural disaster.

Geospatial data for AI training is a booming market, he said. “This is an area of intense focus for us as a business.” Training data is used by government agencies to organize challenges and motivate developers to participate by offering free high-quality satellite and aerial imagery. In three recent challenges, Radiant provided more than 3 million labeled objects in DigitalGlobe satellite imagery made accessible to the public via a “creative commons license.” More public data leads to better open source algorithms, said Frazier.

The company released a free software development kit to help developers apply machine learning models against geospatial intelligence like optical and radar imagery.

“There is a strong desire to take capabilities to teach machines to recognize objects in optical and radar imagery,” said Frazier. “The question is how do I do it at scale across dozens or hundreds of categories of objects?”

The bulk of Radiant’s work is done in unclassified computing environments tapping open source software and commercial data, said the company’s senior director of marketing Andre Kearns. Radiant also employs a large team of cleared developers who are able to retrain open-source models on government specific problems. “The process of retraining these models is called transfer learning,” he said. “The end result is customers benefit from rapid innovation cycles but maintain operational security.”

Radiant is teaming with commercial startups that are trying to break into the national security market, Kearns said. The government benefits from a “vibrant ecosystem” of established companies and newcomers.