On-device machine learning offers a number of benefits ranging from speed to privacy. To accelerate the adoption of machine learning on Android, Google is now adding an “Android machine learning platform” – primarily TensorFlow Lite – to Play services directly.
Google Play services meet the needs of Android users and provide access to tools for 3rd party app developers.
Google found some limitations, preventing apps from leaving machine learning and appearing in the cloud. Google’s solution to this problem is the “Android Machine Learning platform”. Initially, Google directly provides on-device inference on most Android devices.
“TensorFlow Lite will be available on all devices with Google Play Services. Developers will not need to include the runtime in the application, helping to reduce the size of the application “- Google information.
To ensure “optimal” performance across devices, the Automatic Acceleration feature in TensorFlowLite for Android allows “testing each model to whitelist for specific devices taking into account performance, reliability, and performance. accuracy and stability”. The feature will be available later this year and will determine if hardware acceleration is enabled.
Finally, Google will update the Neural Networks API outside of Android OS releases. The company works with chipset vendors, like Qualcomm, to provide the latest device drivers in addition to operating system updates.
“This will allow developers to dramatically reduce testing from thousands of devices to just a few. We’re excited to announce that this feature will be available later this year, and Qualcomm is our first partner,” Google shared.