Dubbed Neuflow, a supercomputer that can function more effectively and swiftly by impersonating the neural network of the human visual system, has been created by Eugenio Culurciello of Yale's School of Engineering & Applied Science. The High Performance Embedded Computing (HPEC) workshop which was held in Boston on September 15 featured his presentation on the outcome of his research.
The intricate vision algorithms utilized by Neuflow helps in operating the large neural networks which can provide the desired artificial vision to the system. These complex algorithms were created by New York University’s Yann LeCun. Both researchers are working on making this system to facilitate autonomous car navigation. By instantaneously processing tens of megapixel images, the Neuflow can identify a variety of things on the road including other cars, side paths, people etc. This system demands a power of a few watts for concurrently performing around 100 billion operations in a second whereas a computer with several graphic processors requires more than 300 W for contributing the same performance.
According to Culurciello, one of their first archetypes of Neuflow performed the vision tasks more effectively than the graphic processors. He implanted the supercomputer on a chip, thereby making it more compact and competent when compared to the usual computers. Culurciello has stated that the size of the entire system will measure the size of a wallet, facilitating its easy incorporation into cars and other systems.
Neuflow could be used to enhance the navigation of the robot in unsafe and inaccessible places. It can also be employed in offering effective artificial vision for defense personnel during warfare, effective monitoring of elderly people who are in assisted living conditions.