Google has introduced TensorFlow Lite, a lighter version of its TensorFlow AI package specially designed to make it easy for developers to deploy machine learning models on mobile and other compact devices.
According to its creators, as it matures, TensorFlow Lite will eventually replace Tensorflow Mobile, the Tensorflow API that already enables model deployment on mobile devices.
At the moment, it has been released as developer preview.
Main features of Tensorflow Lite
According to its creators Tensorflow Lite is:
- Lightweight: Enables inference of machine learning models on the device itself with a small binary size and fast startup/initialization.
- Multiplatform: It is designed to run on many different platforms, starting with Android and iOS.
- Fast: It has been optimized for mobile devices, including significantly improved model load times and support for hardware acceleration.
Advantages
Among other things, Tensorflow Lite will allow you to run in-situ on-device AI algorithms that previously required data to be sent to a server for processing.
This not only speeds up the execution, but also has important improvements in terms of privacy, since there is no need for the user's data to leave the device.
models included
TensorFlow Lite already offers some models that have been trained and optimized for mobile devices:
- MobileNet: models of vision capable of distinguishing between 1000 different object classes, specifically designed to run efficiently on mobile and embedded devices.
- Inception v3: model of image recognition, similar in functionality to MobileNet, which offers higher accuracy but is also larger in size.
- smartreplay: model conversational on the device itself that gives responses to incoming chat messages. Google and third-party messaging apps already use this feature on Android Wear.
However, developers will be able to retrain the models with their own data set if they wish.
In addition, according to the creators of Tensorflow Lite, the offer of models will be expanded later depending on the needs of the users.
For more information, consult the TensorFlow Lite documentation pages.
competition
Google is not the first to bring AI to mobile. Last year, Facebook announced Caffe2Go, a version of Caffe designed for the purpose of running deep learning models on mobile devices.
And throughout this year, products such as Apple Core ML or Clarifai cloud service to train AI models on mobile devices.
Sources:
Keep reading: