Google's vice president of engineering, Dave Burke, announced a new version of Tensorflow optimized for mobile phones.
This new software library, called Tensorflow Lite, allows developers to run artificial intelligence applications on their mobile phones in real time. According to Burke, the library is designed to be faster and smaller, while still supporting the most advanced technology. It will serve asOpen source Tensorflow programPart of the release later this year.
Most of today's artificial intelligence processes servers running on the Software as a Service (SaaS) provider. With the release of the software library, Google hopes to move some of the processing to the user's mobile phone. This not only saves processing power, but also reduces the amount of data. In addition, it ensures that the user's data is private and can no longer rely on the Internet connection.
Tensorflow Lite is the second deep learning tool to run on the mobile phone side. Previously in November 2016,Facebook has released a structure called Caffe2GoThe Caffe2Go for real-time style transfer (Style Transfer), that is, in the user's mobile phone to add a class of art (art-like) filter.
Burke also announced that his team is adding and designing some features for communicating with the neural network processor. Neural networks in the extensive use of the same type of operation, you can use the GPU andGoogles own TPU (Tensor Processing Unit)Optimize for such operations. The current mobile phone has been used for neural networks, for exampleGoogle Translate Appin. For mobile phones, it is difficult to bear the training of such neural networks.