WASHINGTON DC – The U.S. Army’s Office of Business Transformation (OBT) - Office of Enterprise Data Analytics and the Combat Capabilities Development Command (DEVCOM) - Army Research Laboratory (ARL) have concluded the 2022 annual Deep Green Data Science and Artificial Intelligence Competition (Phase I).
The competition’s primary goal is to solve some of the hardest challenges that the United States Army Analytical and Research Departments face today. The second goal is to train the participants in state-of-the-art Deep Learning and Data Science methodologies and technologies, which are designed to increase the ability of the Army's analytical forces.
Beyond the core goals, and above all, the competition is a collaborative learning environment that allows each of the participants to analyze a complex challenge and refine their ability to generate innovative ideas and develop impactful solutions.
2022, Phase I Competition:
- 70 people from the Army, Air Force, and Navy
- 12 teams competed in Phase I of the competition
- 327 models were submitted.
- 108 models met the rigorous constraints and requirements necessary to qualify for final evaluation
Each Phase I model was focused on jump starting the development of an all-terrain autonomous vehicle system. Specifically, each participant worked towards developing an image segmentation model capable of categorizing every pixel in a computer vision digital image.
Out of all the candidates and models, the winning team this year was from the US Army Combat Capabilities Development Command Analysis Center. Their model obtained the highest score by using a common semantic segmentation metric Mean Intersection Over Union (MIOU).
To all the winners, congratulations!
Phase I Winners and Special Awards:
We just finished Deep Greens Phase I and all the winners came from Army Futures Command (AFC)-DEVCOM.
1st Place: Combat Capability Development Command (CDC) DEVCOM (AFC DAC)-Team AFC-DAC
2nd Place: Combat Capability Development Command (CDC) DEVCOM ARL (AFC ARL) – Individual Deep Dreamers
3rd Place: Combat Capability Development Command (CDC) DEVCOM Chemical & Biological Center (Steward Center) (AFC-CBC) – Team CBC-Steward.
The Best in Show Award was given to AFC-DAC for their excellence in model development, methodology, presentation, and overall performance in the event.
The Best in Show Award is given to the team who performed to the highest standards in all aspects of the competition. The team had to produce a model within the top 5 highest scored models. The team also had to present their methodology and results in a clear, concise, and easy to understand presentation.
The Most Innovative Approach was given to Michael Lee for his interesting and unique solution for the challenge.
The Most innovative Approach is given to the team who truly “thought outside of the box” and went above and beyond the materials and approaches taught in the training. The approach not only solved the problem and ranked in the top 5 models, but also showed a unique approach to the solution.
The Most Improved Model was given to AFC-DAC for a model that had the most significant improvement over the course of the competition.
The Most Improved Model was awarded to the team whose difference between their first model submission score to their final submission was greater than the difference from all other teams.
In line with the Secretary of the Army's Strategic Objective #2, Deep Green enables transformation across the Army. It allows the Army's Digital Workforce to exercise their skills at the highest level, and it provides the Army with innovative data science solutions that enable the exploitation of data and the advancement of Army analytics and artificial intelligence capabilities."
According to Mr. Bakari Dale, the Army’s Senior Advisor for Enterprise Data Science & Artificial Intelligence, the Deep Green Challenge is an opportunity for the Office of Business Transformation (OBT) to partner with Army organizations focused on cutting-edge science and technology. “This year, the competition is a unique opportunity for the Army’s data scientists, engineers, statisticians, and operations research analysts. Some of our best and brightest were challenged to tap into their potential and develop innovative machine learning & deep learning solutions that can enable Army autonomous vehicle all-terrain perception. In line with the Secretary of the Army Strategic Objective #2, Deep Green has been transformational for the Army’s Digital Workforce by enabling the implementation of data science and exploiting data for advanced analytics & artificial intelligence use in the Army.”
Upcoming Details of Phase II (2023)
The Deep Green competition is divided into two phases. There are no prerequisites to participate in Phase II, and new teams can register for the competition and get access to training.
On January 9th, 2023, there will be an official kickoff event, with training to follow on Hydrogen Torch from H2O, basic introduction of Python, PyTorch programming, Computer Vision and LiDAR Deep Learning frameworks, and other best practices. Phase II will later conclude on July 28th, 2023, which is the last day to submit models for consideration in the competition.
Phase II of the competition will be broadened in several ways. First, additional data from LiDAR Point-Cloud will be added to the previously available data. This will enable participants to build out robust multi-tiered models.
Second, the competition will make available a no-code solution Hydrogen Torch for the computer vision track provided by H2O.ai. This tool will allow competitors to learn computer vision techniques and processes without the additional burden of learning new code.
Lastly, the competition will open to students in service academies and select military colleges, which will bridge a connection to academia and allow for greater intellectual diversity.
Registration for Phase II is already underway and will continue through March 15th, 2022. To learn more about the competition, or to register, contact Dr. Matthew Miller, Deep Green Project Manager, at firstname.lastname@example.org.