This tiny chip—complete with 12 cores—specializes in enabling low power, advanced computer vision processing which will, no doubt, be a valuable component to “mobilizing” Google’s growing list of products that utilize machine learning.
Machine learning, while complex, is easy to conceptualize—computers and programs that learn and improve over time. They may create their own algorithms to describe their learning, rather than require humans to program them based on known rules.
But what is computer vision?
Simply put, it’s the ability for computing devices to be able to view, interpret and understand images both symbolically and in context. For example, a computer can use a camera to record many images of different species of cats and can catalog them, reference them, compare them and recognize them. No big deal. We do this all the time today.
But to extrapolate beyond just comparing pixels and familiar features and make the jump to understanding the essence of “cat” without a 1 to 1 comparison is a whole different level of computing.
Our human brains do this innately and naturally and can understand complex symbolism (e.g. looking at an abstracted painting of a cat and understanding what it is). Programming this is a different matter altogether.
The Myriad 2 chips, in combination with machine learning, are attempting this sort of operation with a very small footprint and considerable power consumption savings.
In a phone call with Movidius CEO Remi El-Ouazzane, he shared that the power consumption savings of the new chip equates to a 10-100x power savings over current models on the market. It can offer up to five times the device real estate savings over current versions while enabling five times the cost savings.
All of these advantages make it perfect for mobile devices which begs the question “What kind of mobile devices?”
Neither company could comment on exactly what kind of mobile devices will be using the new hardware.
I spoke with Google’s Charina Choi who indicated that the low power consumption of the chip was a key feature in Google’s decision to license the hardware.
She also shared that another big advantage of the chip was that its significant local processing power would enable a host of complex actions to take place resident on devices while they are disconnected from the networks that typically do complex calculations in the cloud. These instantaneous computer vision capabilities will no doubt enable user experiences with low latency on the devices that utilize them.
Blaise Agϋera y Arcas, head of the Google’s machine intelligence group in Seattle further illustrated this by saying that by working with Movidius, they’ll be “able to expand this technology beyond the data center and out into the real world, giving people the benefits of machine intelligence on their personal devices.”