Android 8.1 Introduces Neural Networks API
Written by Lucy Black   
Thursday, 26 October 2017

Google has released the first developer preview of Android 8.1. It includes a new Neural Networks API to accelerate on-device machine intelligence and optimizations for Android Go, as well as smaller enhancements to Oreo resulting from user and developer feedback.


The Neural Networks API (NNAPI) is part of Google's efforts to bring machine intelligence to Android. It is an Android C API designed for running computationally intensive operations by enabling hardware-accelerated inference operations on supported devices. As such it is intended to provide a base layer of functionality for higher-level machine learning frameworks that build and train neural networks.

This is where another piece of news comes in. Google will be bringing us TensorFlow Lite, which it describes as its upcoming cross-platform ML framework for mobile.

NNAPI supports inferencing, such as classifying images, predicting user behavior, and selecting appropriate responses to a search query, by applying data from Android devices to previously trained, developer-defined models. Among its claimed benefits are:


  • Latency: You don’t need to send a request over a network connection and wait for a response. This can be critical for video applications that process successive frames coming from a camera.

  • Availability: The application runs even when outside of network coverage.

  • Speed: New hardware specific to neural networks processing provide significantly faster computation than with general-use CPU alone.

  • Privacy: The data does not leave the device.

  • Cost: No server farm is needed when all the computations are performed on the device.

NNAPI is meant to be called by machine learning libraries, frameworks, and tools that let developers train their models off-device and deploy them on Android devices. Apps typically would not use NNAPI directly, but would instead directly use higher-level machine learning frameworks which in turn could use NNAPI to perform hardware-accelerated inference operations on supported devices.

Its high-level system architecture is given in this diagram:



Other major features relate to Android Go - the version of Oreo, announced at Google I/O in May 2017, that is a lightweight Android capable of operating with only 1 GB, or even 512 MB, of RAM. As well as memory optimizations, Google has added two new hardware feature constants:


to allow devs to target the distribution of apps and APK splits to normal or low-RAM devices running Android 8.1 and later.

Other new features in Android 8.1 include:

  • Autofill enhancements designed to make it easier for password managers and other Autofill services to use the Autofill framework introduced with API level 26. For example, there is added support for more UI customization of the Save dialog, as well as setAutofillOptions() for users to set credit card expiration using a spinner.

  • Shared memory API This new API, which is parcelable, AIDL friendly, and exposes useful features like removing write permissions, lets apps allocate shared memory for faster access to common data. Using it apps can map anonymous shared memory and manage protection controls  The API is parcelable, AIDL friendly, and exposes useful features like removing write permissions.




This initial preview is a beta quality release in which the new APIs are already final and, according to its overview, is suitable for daily use by developers and early adopters. A second preview with near-final system images for final testing will be released in November, with the final release expected in December.


More Information 

Android 8.1 Developer Preview

Neural Networks API

Related Articles

Spectacular Launch For Android Oreo

Android O on the Horizon

Android Instant Apps Updates API

Android Gets Architecture Components

Kotlin Begins Its Takeover Of Android




The University of Tübingen's Self-Driving Cars Course

The recorded lectures and the written material of a course on Self-Driving Cars at the University of Tübingen have been made available for free. It's a first class opportunity to learn the in an [ ... ]

Chainguard Joins Docker Verified Publisher Program

Chainguard has joined the Docker Verified Publisher (DVP) program, meaning its Chainguard Developer Images are now officially available on Docker's container image registry.

More News


raspberry pi books



or email your comment to:

Last Updated ( Wednesday, 15 November 2017 )