Commanders and warfighters in the field rely on the timely processing of data, and the Department of Defense and U.S. intelligence community have an overwhelming data collection capability. But they lack processing power, and most AI/ML techniques are time consuming, expensive, data intensive, and difficult to implement. In many cases, knowledge gained from processed data needs to be quickly available at the tactical edge for decision advantage. What’s more, too few engineers have the expertise to optimize the software for the latest hardware. The result? Decision makers can’t make the most of the data available to them. Principal Investigator Scott McMillan leads a team working on "Spiral AI/ML," an SEI project to automate code generation for data-intensive computations while simultaneously optimizing for targeted hardware. In partnership with Carnegie Mellon University, Scott and his team seek to build a hardware-software co-optimization system that provides for high-performance, data-intensive computing across existing and future) DoD hardware platforms. By providing a standardized high-level interface that allows application developers to develop for any hardware, Spiral promises to speed the development of high-performance AI and ML applications and enable their rapid deployment on new hardware platforms at levels of performance equaling or exceeding hand-tuned software at a fraction of the time and cost.
Read
Spiral/AIML: Resource-Constrained Co-Optimization for High-Performance, Data-Intensive Computing in the online summary booklet
