Artificial Intelligence Stuffed in a Stick

Intel Subsidiary Packs Deep Learning into a Tiny Device

Intel subsidiary Movidius recently launched the Neural Compute Stick. Intel claims it’s the world’s first USB-based deep learning “inference kit and self-contained artificial intelligence accelerator”. A version of the product was showcased at CES 2017.

The Neural Compute Stick adds to Intel’s portfolio of deep learning and artificial intelligence solutions. The concept behind the device is to bring deep learning neural network capabilities to the edge of networks, rather than relying on connections to a central hub or to systems in the cloud.

Intel says the product is designed to enable host devices to process deep neural networks natively, or, in other words, at the edge. In turn, this provides developers and researchers with a low-power and low-cost method of developing and optimising various offline artificial intelligence applications.

According to Intel, the Neural Compute Stick uses a Myriad 2 video processing unit, which provides it with more than 100 gigaflops of performance and consumes only 1 watt. Intel has combined a deep learning inference kit, the means by which artificial intelligence essentially puts into action what it has learnt after initial training, and a self-contained artificial intelligence accelerator to offer dedicated deep learning neural network processing capabilities to a host of devices with a USB connection. The need to constantly transfer data between a device on the edge of a network and a central system becomes redundant, thereby cutting out latency and allowing an artificial intelligence system to take faster decisions and act on them in a short time span.

The compact design of the USB stick allows smart software based on artificial intelligence to be integrated into various products, such as heavy machinery and white goods, paving the way for a high level of smart automation. Intel has been building up other areas, such as server processors, embedded computing and the Internet of things (IoT), to expand beyond the PC market, and artificial intelligence is an increasingly important aspect in those segments.

The Neural Compute Stick is another sign that artificial intelligence is becoming ubiquitous. Like computing power, which has a long history of shrinking in size and price while offering constant exponential boosts in power, now artificial intelligence and deep learning engines are becoming mainstream in a wide range of sectors and applications. In industrial uses, we expect the combination of artificial intelligence and IoT to drive the “software spiral” — the interplay between microprocessors and software that requires evermore computing power — at the edge of the network.