New technology enables U.S. Soldiers to be trained 13 times faster than traditional techniques. According to Army researchers, this may help save lives.
University of Southern California student Shijie Zhou (center) holds a computer board. ARL West’s Dr Raj Kannan (left) and USC Professor Viktor Prasanna (right). (Image credit: U.S. Army Research Laboratory)
Researchers at the
U.S. Army Research Laboratory are enhancing the rate of learning even with scarce resources. Soldiers will be able to decode hints of information more quickly, in turn, speedily deploying solutions, such as identifying potential danger zones from aerial warzone images or threats like a vehicle-borne improvised explosive device.
The scientists used economical, lightweight hardware and executed collaborative filtering, a renowned machine learning technique on a high-tech low-power Field Programmable Gate Array platform to accelerate training by 13.3 times than a high-tech optimized multi-core system and 12.7 times for the optimized GPU systems.
Moreover, the new method considerably reduced the power consumption. The projected consumption was 13.8 watts, as opposed to 235 watts for GPU platforms and 130 watts for the multi-core—thereby making this a potentially valuable component of adaptive, lightweight tactical computing systems.
According to ARL researcher Dr Rajgopal Kannan, in due course, this method could become part of an array of tools embedded on the next generation combat vehicle, providing cognitive services and devices for soldiers in distributed coalition environments.
One of the six Army Modernization Priorities pursued by the laboratory is the development of technology for the next generation combat vehicle.
For this research, Kannan teamed up with a research group at the University of Southern California, namely Professor Viktor Prasanna and students from the data science and architecture lab. Through the West Coast, open campus initiative by ARL, ARL, and USC are endeavoring to speed up and optimize tactical learning applications on heterogeneous economical hardware.
This research is part of Army’s larger focus on artificial intelligence and machine learning research initiatives undertaken to facilitate gaining a strategic upper hand and guarantee soldier dominance with applications such as tactical computing and on-field adaptive processing.
Currently, Kannan is working to develop numerous methods to accelerate AI/ML algorithms through ground-breaking designs on high-tech, low-cost hardware.
According to Kannan, the theoretical techniques can become part of the tool-chain for prospective projects. For instance, a new adaptive processing project that recently commenced, where he is a key researcher, could exploit these abilities.
His paper on accelerating stochastic gradient descent, a method ubiquitous to various machine learning training algorithms, was conferred with the best-paper award at the 26
th ACM/SIGDA International Symposium on Field Programmable Gate Arrays, the premier international conference on technical research in FPGAs, held in Monterey, California, February 25–27, 2018.