The service is especially interested in how it can reconfigure its computational architecture to increase the efficiency of its systems. For algorithms to come to life in neural networks and other AI approaches, engineers need massive computing power to churn through the data. That massive computing power could be realized by more efficient computing structures, and the Air Force wants to know how to do it.
“Unconventional computing architectures are necessary to achieve advanced and new capabilities,” according to the Broad Agency Announcement.
The Air Force Research Laboratory will review submitted white papers, offering between $1 million and $3 million for effective solutions.
Some of the ideas the Air Force is toying with to better use its computing systems are modeling them after “brain-inspired computing architectures.” These architectures are known as “neuromorphic” and were developed in the late 1980s to design analog circuits that imitate how synapses in the brain transmit information.
The lab’s intent is to develop innovative modular computing systems, potentially like neuromorphic computing, to meet increased future demand for data bandwidth. More data usually means more accurate AI.
This type of research has been championed by outside groups that have urged the department to invest in foundational research on AI. The white papers the lab is requesting will be used to improve the Defense Department’s future fielding of AI.
Air Force and other IT leaders have said that the department is in its first, “foundational” phase of collecting and governing data more effectively to turn that data into the fuel that AI algorithms can run on. So far, data in the department has been scattered and difficult to use properly without uniform cloud-based computing systems, the officials have said.
The call for white papers will be open for a long time. The Air Force has set aside $99 million to fund papers through September 2023.
“The overarching objective is to achieve orders of magnitude improvement in size, weight and power for deploying robust artificial intelligence and machine learning (AI/ML) capabilities in an embedded computing environment,” the BAA states.