The U.S. Air Force Research Lab (AFRL) is exploring whether brain-inspired computer chips could give satellites, aircraft, and drones the ability to automatically identify vehicles.
IBM's neuromorphic TrueNorth chip has successfully detected military and civilian vehicles in radar-generated aerial imagery while using less energy than a regular high-performance computer. The chip processes data using a network of elements designed to mimic neurons and synapses that store and operate on data, making the chip more efficient than conventional systems, in which components that perform calculations are separate from memory.
AFRL conducted contests between TrueNorth and the Jetson TX-1, a high-powered NVIDIA computer. The systems used different implementations of neural network-based image-processing software to identify 10 types of military and civilian vehicles. Both systems achieved about 95% accuracy, but the TrueNorth chip used between 3% and 5% of the power required by the NVIDIA computer. The conventional computer ran its neural network on chips with general-purpose hardware, while the IBM chip's hardware is hard-coded to represent artificial neural networks using 1 million neurons customized to the task.
However, it is much easier to deploy neural networks on conventional computers due to a wider availability of software, and the IBM chip is much more expensive.
From Technology Review
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found