As AI technology grows in popularity and complexity, the natural progression is to shrink the actual size this type of computing power needs to function, and that is exactly what a Silicon Valley chip designer has done.
Chip designer Movidius has launched a USB stick with a supercomputer onboard, essentially taking the deep learning paradigm of artificial intelligence to the spatial efficiency of a USB flash drive. Deep learning involves training a computer which, as we saw with Google’s AlphaGo, can create a true force to be reckoned with. Dubbed the “Fathom Neural Compute Stick,” the device has been designed to connect to existing systems and increase the performance of deep learning tasks by 20-30 times.
The chip maker is applying this leap in technology to other endeavors as well, such as helping drones avoid obstacles and developing higher sensitivity thermal imaging. The company has also signed a deal with Google which may take their DeepMind project to new heights.
Facebook’s director of artificial intelligence, Yann LeCun, states: “As a participant in the deep learning ecosystem, I have been hoping for a long time that something like Fathom would become available. The Fathom Neural Compute Stick is a compact, low-power convolution net accelerator for embedded applications that is quite unique.”
Containing the same chips used in high-value drones and professional cameras, the Myriad 2 chip can handle astronomical amounts of processes simultaneously while consuming no more than 1.2 watts.
Possibly the most impressive feature of Fathom is its price tag. Not only is it promising for such a powerful tool to become available in a pocket-sized medium, but with a price tag of just under $100, the audience of potentially world-changing users becomes virtually limitless.
Finally, with advanced VR headsets and titles being released at shorter intervals, Fathom could augment detail and further reduce the price tags of newer devices on the market. Fathom is an excellent milestone for the size-to-efficiency ratio of AI processing and, while it is not immediately hailing some Matrix-style robot overlords, steps toward computerized companions can always be considered impressive.
