Google's new deal with chipmaker Movdius will develop more advanced machine intelligence processing on mobile devices such as Android handsets, to produce real-time facial recognition. The search giant has already squeezed a neural network into a smartphone through its Google Translate app, which uses the phone's camera to capture images allowing the processor to translate text.
SwiftKey also uses a small-scale neural network. It is for word predictions in the SwiftKey Neural application.
Google has already used Movidius' Myriad 1 visual-processing unit (VPU) chip in developing Project Tango devices in 2014. It improved the Android phones' 3D-mapping capabilities.
The search giant is now licensing the company's newest technology to add deep learning capabilities to mobile devices, according to Engadget. This involves a "neural network" computer program that functions like a human brain.
Movidius announced that Google will use the company's most powerful chip from the Myriad 2 family of VPUs to perform various tasks. They include recognizing people, objects, and text, according to ZDNet.
Blaise Agϋera y Arcas is head of Google's machine intelligence group. He explained that a traditional microprocessor functions on data in a series, but Myriad operates in "torrents" of information that all enters at one time.
Arcas shared that Google's teaming up with Movidius will move the tech out from datacenters. It will allow people to benefit from machine intelligence on their devices such as smartphones, tablets, and PCs.
Neural networks have been key in improving upgrades for image and speech recognition tools. This process usually require an Internet connection, such as when an Android phone processes the user's voice.
Project Tango is already dealing with one possible problem. It is the powerful Myriad chip's effect on smartphone batteries that are already being pushed to their limits.
Movidius CEO Remi El-Ouazzane believes they have found a solution. He told CNET that battery power will be saved such as by processing data captured by the smartphone's camera.
El-Ouazzane highlighted the importance of using sky-high power efficiency. It could be achieved through the mobile phone's hardware architecture and neural computing.
Movidius is an image-processing company that was founded in Ireland in 2005. Several companies doing work in computer vision have teamed up with the company.
Here's a demo of Google's Translate app: