Qualcomm's Deep Learning Tool To Make Mobile Devices Smarter, More Secure
A new software development kit from Qualcomm aims to bring machine-learning capabilities to smartphones and other mobile devices powered by its chips. The SDK, known as the Qualcomm Snapdragon Neural Processing Engine, will allow original equipment manufacturers to design and run their own neural network models on devices powered by the company’s Snapdragon 820 chip, without requiring a connection to the cloud.
The SDK will also represent the first deep learning toolkit optimized for a mobile system-on-a-chip (SoC) platform such as Snapdragon. That could potentially open up a range of new capabilities for mobile devices, including scene detection, text recognition, object tracking and avoidance, gesturing, face recognition and natural language processing, according to the company.
On-Device Deep Learning
“The demand for untethered, mobile learning-driven user experiences is increasing rapidly and Qualcomm Technologies’ customers have been clamoring for tools to help them realize their product visions,” Gary Brotman, Qualcomm Technologies' director of product management, said in a statement. “With the introduction of the new Snapdragon Neural Processing Engine SDK, we are making it possible for myriad sectors, including mobile, IoT and automotive to harnesses the power of Qualcomm Snapdragon 820 and make high-performance, power efficient on-device deep learning a reality.”
The new SDK will be powered by Qualcomm’s Zeroth Machine Intelligence Platform, a machine intelligence software platform specifically optimized for the Snapdragon SoC. The same technology currently powers visual intelligence software such as Snapdragon Scene Detect and advanced malware detection software found in Snapdragon Smart Protect.
The SDK was announced earlier today at the Embedded Vision Summit in Santa Clara, Calif. In addition to bringing deep learning to smart phones, the company said the SDK will also make it possible to provide similar capabilities to a variety of devices such as security cameras, automobiles, and drones.
Optimized for Healthcare, Automotive, Security, Imaging
The SDK will also include a number of other advanced features, including accelerated runtime for on-device execution of convolutional and recurrent neural networks on the Snapdragon 820 cores, such as the Kryo CPU, Adreno GPU, and Hexagon DSP, according to Qualcomm. In addition, the SDK will provide support for common deep learning model frameworks, such as Caffe and CudaConvNet.
Qualcomm said the toolkit is both lightweight and flexible, and designed to take advantage of Snapdragon’s heterogeneous cores to drive improved performance and reduce power consumption for mobile devices. It is also geared toward providing support for companies in a broad range of industries, including healthcare, automotive, security and imaging, to run their own proprietary trained neural network models on portable devices.
“The Neural Processing Engine SDK means we can quickly deploy our proprietary deep learning algorithms to our Snapdragon-based connected camera devices in the field, which can detect driver distraction and help prevent auto accidents,” said Frederick Soo, chief technology officer of NAUTO, an autonomous vehicle technology company, in the statement.
Qualcomm said the SDK is expected to be available for Snapdragon 820 processors in the second half of 2016.
Image Credit: Snapdragon phone and processor (pictured above) via Qualcomm.
Image credit: Product shots by Qualcomm.