DARPA Funding Announced for Professor Kaeli
$753K Grant Addresses High-Performance Computing for DARPA
COE Distinguished Professor David Kaeli from the Department of Electrical and Computer Engineering has received a four-year, $753K Software-Defined Hardware TA-2 grant from the Defense Advanced Research Projects Agency (DARPA).
Much of today’s work in big data uses complex languages that would typically require super computer facilities. This grant—titled “Mitchell”—funds a high-performance computing project at the intersection of hardware and software in which Kaeli and his team seek to create a domain-specific language that will be more natural for data scientists to work in that can efficiently leverage a very high-performance system created by Intel.
“This funding from DARPA is an effort for the federal government and the Department of Defense to allow for reconfigurable hardware that can adapt based on data intensive applications,” says Kaeli. “This state-of-the-art language will allow users to employ Intel’s hardware more simply on computationally challenging applications such as machine learning, artificial intelligence, and graph analytics.”
For example, training an autonomous vehicle to identify “good guys” versus “bad guys” requires millions of examples for the machine to learn the difference between the two. This project would increase the ability for machines to learn complex frameworks more efficiently, but without giving up performance.
The grant is part of DARPA’s Electronics Resurgence Initiative (ERI), which selected a number of research teams from both academia and industry in summer of 2018 to “explore the development of flexible architectures capable of using specialized hardware to solve specific computing problems more quickly and efficiently.”
“Mitchell” partner and prime investigator Systems Technology Research—a defense contractor in Woburn, MA—reached out to Kaeli based on his previous work in high-performance parallel systems in compilers, runtimes, and optimization. Purdue University is another partner on the grant, bringing their expertise to produce a domain-specific language that Kaeli and his team will then map to Intel’s architecture.
While the DARPA grant is aimed at a very specific goal for very specific hardware, much of this technology will eventually end up in the public sector. High-performance computing will impact programmers’ ability to do highly efficient cloud computing on complex large data sets in a language they’re more comfortable with.
“There is currently a gap between high-performance systems and programming frameworks, and we’re looking to improve that,” says Kaeli. “Intel builds systems that are made to be used by everyone, so creating a more simplified language that works with their hardware on large data sets will have important ramifications on the future of the use of big data in society.”