close
AI

Focus of Google Developers Conference: New head of Greater China debut, Android 10 native support 5G, TF2.0 big update

39748706-d545-11e9-ba9a-aaaa00151f43.jpg

Editor’s Note: This article comes from WeChat public account “New Intelligence” (ID: AI_era), author Peng Fei, Daming, Zhang Jia, 36 氪 authorized to release.[New Zhiyuan Guide] The two-day 2019 Google Developer Conference has just come to an end. Google has launched the latest Android 10, TensorFlow 2.0, Flutter 1.9 and other products and development tools.In addition, Chen Junting, the new president of Google Greater China, debuted and announced the official launch of the Grow with Google Growth Plan.In 2016, Google Developers Conference was held for the first time in China. In 2017, Google announced the establishment of “AI China R&D Center”. In 2018, Google’s “AI+ Small Program” product, “Guessing Small Songs”, was popular in China, this year’s Google Developers Conference.What brought us?The 2019 Google Developer Days (GDD) has just come to an end. During the two-day conference, Google launched a number of new products and development tools: the Android team updated the latest product Android 10, and the TensorFlow team released TensorFlow.2.0, Flutter launched the Flutter 1.9, Web, Firebase, Wear OS and ARCore teams also introduced related product updates.In addition, Chen Junting, the new president of Google Greater China, announced the official launch of the Grow with Google Growth Program, which will offer a range of free online courses, learning tools and offline training to help more people unlock the essential skills of the digital age.Let’s take a look at the highlights of this Google Developer Conference.Chen Junting, the new president of Google Greater China, first appeared: launching Grow with Google’s growth plan In the Keynote keynote session, the new Google Greater China President Stanley Chen first appeared.This is the first time Chen Junting appeared in the public meeting after taking up the position in June this year.He announced Google’s data technology sharing program, which has benefited more than 43 million people in more than 80 countries and regions.After reviewing some of Google’s recent moves, it announced the official launch of the Grow with Google Growth Program, which provides free course learning programs to developers.Chen Junting also said in his speech that he hopes to provide free courses through the online sharing platform, bringing Google’s technology to more people who need it, helping users to use and experience Google’s products.Chen Junting graduated from the National Taiwan University Economics Department and joined Google China Taiwan in 2011 as the general manager of Taiwan.Responsible for sales business, responsibilities covering large enterprise customers, distribution and online business.Prior to joining Google, Chen Junting worked for Apple as a business associate in the Taiwan branch of China and head of campus business in Asia.Android 10 heavyweight debut, native support 5G, more attention to data security This conference officially launched a new version of the Android operating system Android 10.In fact, a week before this developer conference, Google senior Android product manager Stephanie Cuthbertson has announced the news on the Android developer public number.Cuthbertson said the new Android 10 focuses on mobile innovation, security privacy and digital health to create the best user experience.In mobile innovation, several highlights of Android 10 include: strong support for 5G networks: Android10 provides 5G platform support, and extends existing APIs to better help developers use 5G technology to drive application innovation.Smart reply in the notification: Using the built-in machine learning technology of the device, the recommended action related to the current scene is displayed in the notification, such as a message smart reply, or when the notification includes the address information, the user is suggested to open the map application.The smart reply feature provides suggestions to the user based on the notification content. The official dark theme: Android 10 adds a system global dark theme, allowing users to use the device more easily in low-light environments and greatly reducing device power consumption.the amount.Google Keep gesture navigation in a dark theme: When enabled, the original bottom navigation bar design disappears, allowing the app to actually display full screen.Gesture navigation allows applications to display full-screen content in terms of privacy protection, data security and network performance. Android 10 also makes a number of improvements and updates, mainly in the following aspects: giving users more geographic control: in order to improve user controllability,Users can access device location information only by the application being used (running in the foreground) with a new geographic location.Prevent device tracking: In order to prevent application tracking devices, Android10 does not allow general applications to access non-resetable device identification numbers such as IMEI codes and serial numbers.P2P and network connectivity optimization: In Android 10, we refactored the Wi-Fi stack, which not only improved the performance of the application in terms of privacy and performance, but also optimized for many common use cases, Wi-Fi performance mode: Android10 allows applications to enable high-performance, low-latency adaptive Wi-Fi mode, so that even in scenarios where network latency is sensitive, such as real-time gaming, audio calls, etc., applications can still guarantee a smooth user experience.In performance mode, the Android platform works with device firmware to achieve the lowest power consumption.TensorFlow 2.0, more intuitive, more powerful and more flexible. This GDD, TensorFlow 2.0 is still the highlight.TensorFlow Asia Pacific Product Manager Liang Xinping started sharing “The Future and Future of Machine Learning” and revealed the current home of TensorFlow: since its release in 2015, it has reached 41 million downloads, more than 50,000 submissions, 9900Subcode change request and more than 1800 contributors.At present, a huge ecosystem has been built around TensorFlow, and TensorFlow has become the most successful machine learning platform.The TensorFlow 2.0 code is very similar to NumPy, and the two can easily replace objects with each other.Developers don’t have to worry about placeholders, Sessions, and feed_dictionaties.The 2.0 release has been upgraded in three areas: ease of use, high performance, and scalability.Below we list some of the main upgrade aspects: Instant execution (Eager Execution) becomes the default mode 1.0 era to create a neural network code as follows: But by the 2.0 era it has become: API cleanup and primary API like TensorFlow.gans, TensorFlow.appMany APIs, such as TensorFlow.contrib and TensorFlow.flags, are either removed directly or moved to a separate library.There is more than one way to create and train ML models in TensorFlow 1.x.Most of the time, developers are not very clear about which API should be used under what circumstances.And once you’ve used one of the API training models, it’s hard to switch to another API.So in TensorFlow 2.0, TensorFlow.keras is the recommended high-level API.The primary API from TensorFlow 1.x to 2.0 did not change much.When installing TensorFlow via pip, you will generally get the full Keras API and some additional set of function tools.Defining Training Cycles In TensorFlow 2.0, the easiest way to train a model is to use the fit() method, which supports both serialization and subclassing models.Developers can also easily configure a TensorBoard instance via the fit() method.It also applies to Jupyter/Colab notebooks.When using model subclassing, the only thing you need to do is override the class member function compute_output_shape() or simply discard it.In addition, developers can use TensorFlow.data.Dataset or NumPy’s standard nd-arrays as input to fit().Model Saving and Restoring The criteria for how to save and load models for products in TensorFlow 1.x are missing.TensorFlow 2.0 attempts to solve this problem by defining a single API.In addition, the TensorFlow ecosystem integrates SavedModels so that it can be deployed to mobile phones, edge devices, servers, and more.TF.Text: Training the NLP model TF.Text is a TensorFlow 2.0 library that can be easily installed using PIP commands.Ability to periodically perform pre-processing in a text-based model and provide more features and operations on language modeling not available in the TensorFlow core components, providing powerful text processing for TensorFlow 2.0 and compatibility with dynamic graphsmode.TensorFlow Lite: Deploying Machine Learning TensorFlow Lite on the End For mobile phones and embedded devices, the framework for deploying machine learning applications is characterized by virtually no delay, no need to connect to the network, and privacy protection.For example, in the rental scene, TensorFlow Lite automatically labels the pictures to improve the efficiency of renting; Cobos robot deploys TensorFlow Lite in the sweeping robot to achieve automatic obstacle avoidance and so on.TensorFlow Lite is also widely used in Google products, such as Google Photos, input methods, cloud assistants, and more.According to statistics, there are more than 2 billion mobile devices, and TensorFlow Lite-based applications have been installed.TensorFlow.js: A platform for making WeChat applets TensorFlow.js is a deep learning platform customized for JavaScript.You can run existing models, retrain existing models, and train new models to support multiple platforms: browsers, wireless (such as WeChat applets), servers, desktops.In addition to running machine learning models on multiple platforms, TensorFlow.js also trains models, has GPU acceleration, and automatically supports WebGL.The engineer demonstrated the smallest and fastest virtual makeup test using TF.js.It is reported that the follow-up will also achieve haircut conversion, age conversion simulation, skin quality detection and other functions.Flutter 1.9 stable release, support for macOS Catalina and iOS 13Flutter is a UI toolkit from Google that allows developers to build beautiful applications for mobile, web and desktop with just one set of code.According to reports, Flutter 1.9 is the largest version update to date, with more than 100 contributors submitting more than 1500 pull requests during this development cycle.The new version has merged Flutter for web into the main Flutter repository and supports macOS Catalina and iOS 13, improved toolchain, support for new Dart language features and new material widgets..