DARPA Announces Winners of Artificial Intelligence Competition to Aid Critical Minerals Assessments
Winning solutions will help the USGS automate key steps in evaluating geologic maps of mineral deposits that are vital to the U.S. Economy
RESTON, Va. — Critical minerals are essential to the U.S. economy and national security; however, their supply is vulnerable to disruption. U.S. production and refining of critical minerals has been declining for decades, while production has become more concentrated in fewer countries.
Given the urgency to increase and better secure critical-mineral supply, The Defense Advanced Research Projects Agency (DARPA) partnered with the U.S. Geological Survey (USGS) to launch the Artificial Intelligence for Critical Mineral Assessment Competition in August 2022.
The partnership will help the USGS conduct more than 50 assessments of critical-mineral resources to aid in economic planning and land-use decision-making. To do this, the USGS draws from more than a century of accumulated data, contained mostly within geologic maps and reports, that provide the fundamental basis for these resource assessments.
Extracting useful and accurate information from these maps is a time-consuming and laborious process involving manual human effort. In fact, a typical assessment for one critical mineral takes approximately two years to prepare. That’s because the USGS map catalog consists of more than 100,000 geologic maps; only about 10% of those are available as georeferenced images and only about half of those are fully digitized vector files needed for analysis. Everything else – 90% of the data – consists of scanned images of paper maps.
The goal of the competition was to crowdsource ideas that could drastically reduce the time required to complete parts of the assessment, using AI and machine learning to automate key processes.
“The competition has been a valuable opportunity for the USGS to work with leading minds in AI to improve our approach to critical-mineral assessments,” said David Applegate, USGS Director. “It has already led to incredible time savings in how we prepare data in a machine-readable format. Furthermore, these machine-learning models have implications beyond mineral resources into other fields that use map data, including geologic mapping, ecological mapping of species diversity and many other application areas.”
“We anticipate our experience will serve as a road map for future interagency collaborations where machine learning can be applied to real-world problems,” said Bart Russell, deputy director of DARPA’s Defense Sciences Office.
After analyzing the mineral-assessment workflow, DARPA and its performers MITRE and NASA Jet Propulsion Laboratory recognized the greatest potential for near-term, high impact was in solving the data needs associated with georeferencing and extraction of individual geologic features found on USGS maps. As such, the competition was divided into two distinct sub-challenges. A total of 18 teams from industry, academia and even a high-school junior competed for cash prizes of $10,000 for first place, $3,000 for second and $1,000 for third.
For the Map Georeferencing Challenge, participants were tasked to find a map within a given scanned image and georeference it by aligning reference points to base maps, such as grid lines, topography, administrative boundaries, roads, or towns. A Canadian company, Uncharted, received top prize for their simple, clean and organized solution. U.S. company Jataware received second place, and “Team Ptolemy,” with members from the Massachusetts Institute of Technology, University of Arizona and Pennsylvania State University, received third place.
For the Map Feature Extraction Challenge, participants were asked to extract features identified in an image’s map legend. Students and faculty from the University of Southern California Information Sciences Institute and University of Minnesota joined forces, earning first place for their exceptional solution to extract line features as well as polygons and points. “Team ICM” from the University of Illinois received second place, followed by Uncharted in third.
Throughout the competition, participants had up to eight weeks to complete each challenge. Each week, they had the option to submit their results for a blind validation dataset to test the accuracy of their code. In the last week of each challenge, participants received a completely blind evaluation dataset and had 24 hours to process and submit their code and detailed documentation of their approach, which was evaluated by experts from USGS, MITRE and NASA Jet Propulsion Laboratory, who reviewed the solutions for accuracy/usability.
To meet the high quality standards required by the USGS, the resulting solutions require further evaluation and development to become operational. USGS experts plan to integrate the best elements of the submissions into a workable solution for mineral assessment workflows and potentially for other mission area assessments within the agency.
In addition to identifying fresh approaches for this problem, DARPA officials view these competitions as a model for how transition partners can access the agency’s performer base.
To hear more about the competition, including insights from members of the winning teams, listen to Voices from DARPA podcast episode 63, “So Many Maps, So Little Time: Using AI to Locate Critical Minerals.” A list of winners can be found on the DARPA website.
Get Our News
These items are in the RSS feed format (Really Simple Syndication) based on categories such as topics, locations, and more. You can install and RSS reader browser extension, software, or use a third-party service to receive immediate news updates depending on the feed that you have added. If you click the feed links below, they may look strange because they are simply XML code. An RSS reader can easily read this code and push out a notification to you when something new is posted to our site.