More and more applications make up the machine learning on the device itself to offer lower latency in predictions, more efficient use of the battery and most importantly not to depend on a network connection, but it also involves some problems that Google now wants to solve.
Google has found that developers who implement machine learning in their applications face application size limitations, performance differences, and limitations in the latest advances in machine learning. To solve these three problems Google will integrate TensorFlow Lite machine learning in Android.
TensorFlow Lite for Android
Currently, most applications that use machine learning integrate the library TensorFlow Lite in its APK by increasing the size of the application, with which Google has decided integrate TensorFlow Lite directly into Google Play services.
TensorFlow Lite integration on all compatible mobiles prevents applications from having to integrate this library, slightly reduce the size of applications, but it also brings two other important novelties.
This integration will also allow Google improve optimal performance on all devices in your library. On some devices, machine learning may enable the hardware acceleration when it is available for certain AI tasks to complete in less time.
By integrating the TensorFlow Lite API into Google Play services, you will receive regular updates on all versions of Android 4.4 or higher, thus covering practically all active devices on the market. In addition, Google is also working with chip vendors to update hardware drivers through Google Play, with Qualcomm being the first partner.
The integration of TensorFlow Lite into Google Play services will arrive on Android by the end of the year. Starting today, developers can sign up for the preview to test machine learning integration on Android.
Via | Android Developers
