 "Google announces TensorFlow Lite developer preview, bringing artificial intelligence capabilities to mobile and embedded devices")
Google has announced the developer preview for TensorFlow Lite, a version of Google’s **TensorFlow** open source library primarily used for machine learning applications. TensorFlow Lite was first announced by Dave Burke, VP of engineering of Android during Google I/O 2017 . [caption id=“attachment_4210975” align=“alignleft” width=“380”]  Image: Tech2[/caption] The idea was to allow developers to create applications that would use on-device machine learning capabilities the same way as Google does with Android. However, TensorFlow Lite has been designed to be cross-platform, and developers can create applications for both iOS and Android. TensorFlow Lite has a low memory footprint to make it less taxing on limited resources available on mobile devices, as well as less processor intensive to make the applications fast. TensorFlow Lite is seen as the replacement for TensorFlow Mobile, but considering the preview status, TensorFlow Mobile is the current choice for stable applications. TensorFlow Mobile allows developers to integrate TensorFlow models that work in a desktop environment, on mobile devices. However, applications created using TensorFlow Lite will be lighter and faster than similar applications that use TensorFlow Mobile. Not all use cases are supported by TensorFlow Lite at the moment though. There are three models that are already trained and optimised for mobile devices. MobileNet and Inception V3 are image recognition models. MobileNet is smaller, but Inception V3 is more accurate. Smart Reply is a model that generates automatic replies to conversational text messages. There are no pre-trained models at the moment for voice synthesis, translation speech recognition, all of which are available on TensorFlow Mobile. While you can process new data on the device through applications based on TensorFlow Lite, at this point, it is not possible to actually train new machine learning models on the device itself. In its current form, TensorFlow Lite supports inference and not training.
We are so excited to announce the Developer Preview of #TensorFlowLite!
— TensorFlow (@TensorFlow) November 14, 2017
Read all about it in our blogpost → https://t.co/OhP8Z9dEiO pic.twitter.com/mI61bvD6Vm
TensorFlow developers have indicated at some enhancements that will be available down the line, especially the support for new models. There are expected to be performance improvements, introduction of simplified tools for developers, and support for smaller devices. At some point, mobile devices are expected to get capabilities for on-device training as well. The TensorFlow Lite Developer Preview is available on GitHub, along with code samples and demo applications.
](https://images.firstpost.com/wp-content/uploads/2017/11/tensorflow-lite-380.jpg){kind=link}