This article is more than 1 year old
Qualcomm goes for the grey matter with neural net SDK
Braaaaaiiiins. Fat, rich, digital braaaaaaiiiins
Qualcomm wants to stick a neural network in your hands, and has announced it will ship a software development kit (SDK) in the second half of this year to get the ball rolling.
The deep learning SDK for the Snapdragon 820 processor will be, the company says, designed to take advantage of the SoC's “heterogenous compute capabilities”.
It's got a snappy name, too: the Qualcomm Snapdragon Neural Processing Engine, powered by the company's Zeroth Machine Intelligence Platform (waits for audience applause, as usual, not a sausage).
What the developer gets is a place to develop neural network models, and compile them to run on the Snapdragon 820. Application examples the company's thought of so far include “scene detection, text recognition, object tracking and avoidance, gesturing, face recognition, and natural language processing”.
The company has already used the Zeroth software platform in its own projects, it says, such as Snapdragon Scene Detect and the Smart Protect malware detection system.
Deep learning frameworks supported by the SDK include:
- The Caffe framework, developed by the Berkeley Vision and Learning Center, a high-speed environment targeting vision systems; and
- Cuda ConvNet, a high-performance C++/CUDA implementation of convolutional neural networks.
Interestingly, Qualcomm reckon neural network models can run just as happily at the device level, without having to offload processing to the cloud.
To that end, it says the SDK is designed to balance power and performance in the Snapdragon 820, when running deep learning software. ®