WASHINGTON DC – Over the past year, the Office of Enterprise Management (OEM) has worked alongside the US Army Combat Capabilities Development Command Army Research Laboratory (DEVCOM ARL) on the Army’s premiere Data Science and Artificial Intelligence Competition: Deep Green. Deep Green is an open DoD competition with two primary goals; one is solving the hardest data driven challenges across the Army and DoD. The second goal is to train members of the DoD in state-of-the-art tools, methodologies, and to fully upskill the Army and members of the DoD in advanced AI approaches. The Deep Green challenges have been vastly successful over the past 4 years, yielding a production ready solution for each sponsoring office. This year’s challenge, sponsored by DEVCOM ARL, was to build a semantic segmentation model for use in their autonomous all-terrain vehicle systems. Semantic segmentation is a challenging computer vision application which classifies each pixel of an image to a specific class. These models are common in autonomous real-world systems requiring full environmental understanding.
DEVCOM ARL and OEM would like to congratulate the members of the AI Avengers team from the Army Communications Electronics Command (CECOM) in winning the 2023 Deep Green Artificial Intelligence and Machine Learning Challenge! The team consists of Anthony Frampton, Frank Frisby, Montgomery Maluki, and Roy Trieu. Their winning model outperformed others by approximately 6 percent.
Mr. Bakari Dale, the Business Mission Area Data Office and Army Senior Leader in Enterprise Data Science and AI states that: "The Deep Green competition is not just a contest; it's a catalyst for transformative change within the Army and the DoD. It's been incredible to witness how the competition has become a powerful platform for professional development, upskilling our soldiers and civilians in advanced AI and machine learning techniques. The triumph of the AI Avengers team exemplifies the potential of this combination of commitment and technology. Their innovative approach to solving a real-world Army computer vision problem underscores the strides we're making in harnessing the power of AI to enhance our operational effectiveness. This victory is a testament to the team's prowess and a reflection of how far we've come in embracing advanced AI approaches. Well done and thank you to all the contributors supporting Deep Green and our efforts to strengthen and advance our Army and the DoD."
Since Deep Green started in 2019, OEM has trained over 200 individuals across multiple domains ranging from financial management, preventative maintenance, and advanced computer vision. By working in this cross-domain area, Deep Green offers a great opportunity for participants to learn and engage in cross-skill training opportunities leveraging machine learning and deep learning. The coded models produced by competitors provided solid performance, however, the model structure provided by the winning team was able to marry the strength of the software with the team’s innovative approach. Philip Osteen, the Army Research Lab competition sponsor comments that “the no-code tool facilitated rapid iteration on architecture design, data processing, and training hyper-parameters, which in the end produced superior models. The coded models were also innovative, particularly those leveraging depth information, and provided strong competitive results. Using the flexibility of coded approaches with the rapid iteration capabilities of no-code solutions is a promising direction for future work.”
OEM is constantly on the lookout for new problems to solve using Deep Green, if your organization has a challenging data problem that you need help with use this link to submit a use case. If you are interested in participating as a competitor in a future Deep Green challenge keep an eye out here for the posting of details for the next challenge. The competition is open to the DoD, the Intelligence Community, other US Government Agencies, or member of a State or Private US based University (with prior approval).
Social Sharing