Developer Keynote (Google I/O '19)

♪ (music) ♪ Good afternoon, everyone Welcome to I/O! (cheers and applause) For about a third of you here live, this is your first I/O And it’s mine, too (cheers and applause) I’m so glad to be here with you, live, and to welcome all of those on livestream You know, we have developers here from 120 countries around the world, and many of you watching us from around the world Thank you for taking the time (cheers and applause) Thank you for making time, no matter what time it is where you are, to invest in yourself, your knowledge, and to spend time with us as we tell you what we’re super excited about showing you today Many of you are connected through our developer community groups You know, we have almost 1,000 Google developer groups around the world And if you’re one of our Google developer experts, each of you do so much to help your local developer community learn and grow with Google We appreciate everything you do for us (cheers and applause) You know, we each have our own unique paths to becoming developers Some of you here are students, and just by being here you’ve got a leg up on your paths to learning When I was in college, my first job was in the Computer Science Lab, learning to program the first 3D graphical visualization library for gaming and entertainment software on a Silicon Graphics IRIS workstation Developing that visualization library was so magical, because people could create amazing special effects with it so quickly And, even though I’d never programmed before then, I felt confident that, with the cutting-edge tools and some helpful friends, I could do it, I could figure it out And that’s the feeling we want each of you to have today, and every day, that anything is possible, and that you can create that next amazing experience, app, or startup There’s never been a better time to be a developer There are three important reasons that I believe this is true First, new user interfaces such as voice, camera, speech, are making computing ambient, available to humans wherever they are, all the time, and enabling much richer forms of interaction than were possible through keyboards and mice Second, as you heard this morning, the Assistant provides great facilities to make computing much more helpful and convenient for people And third, with the advent of cloud computing, we offer incredibly effective tools for developers, whether that’s the convenience of serverless and container technology or the ability to infuse artificial intelligence and machine learning into your applications Now, Google was founded by developers, and we’re made up of engineers and developers Our engineers work hard to bring you new platforms and APIs so that you can get your idea and your passion in front of billions of users, including the next billion to come from around the world We give you an opportunity to see how far your ideas can go, and we’re grateful to play a role in this With this opportunity also comes responsibility Each of us shares a responsibility to our users to make good things for them, to be advocates for them, to respect them, to help them, to delight them with what you build with your creativity and passion We at Google have a responsibility to you, the developer community We seek to empower you with the technology and tools to help you build, grow, to learn, and to earn And that’s exactly what we’re here to do today

We’re going to spend the next hour sharing an awesome set of updates from our developer teams at Google, who are working to enable and support you on your unique path, whether that’s the latest advances in our Assistant, machine learning, Android, web, and more We have a fantastic show planned for you To get started, I’d like to welcome Chet to share the latest from Android ♪ (music) ♪ (cheers and applause) We heard in the earlier keynote about the amazing ecosystem that Android has become I want to focus on two aspects of that ecosystem One is that Android is a product that brings rich experiences to users– that’s all of you– all over the world The other is that Android is also a platform that brings capabilities to developers– that’s also all of you– to build powerful applications So today I’m going to talk about the things that we’re doing to ensure the users can have a product that they trust while developers become more productive with our platform So, first, let’s talk about responsibility to our users As a developer community, we all care about getting this right It’s about building a platform that gives developers powerful capabilities while making sure that users feel that their safety and privacy is protected When we released Android Q Beta a few months ago, we had nearly 50 features and improvements that were all about giving users more control and transparency over their experience But some of those changes required more effort from you, the developers for the platform A good example of that is the changes in storage We got strong feedback from you in the Beta 1 and Beta 2 releases that helped us find a better way So you’ll see in the Beta 2.5 and 3 releases that we have a new approach that reflects the input that we got from you and makes it easier to adopt to those changes going forward So thank you for the feedback, and please keep it coming This is actually exactly why we do preview releases, because we want to iterate with you to find the best solution for everybody before the final release It’s why we do previews, and it’s also why we love our developer community Speaking of developers, let’s talk about developer productivity We care about productivity for developers In fact, we care deeply about productivity for developers, because we are all developers, as well And we know that the more productive we are, the more code we get to write So, two years ago, we announced Kotlin as a supported language for Android Many developers were using Kotlin already, and many more have been using it since In fact, over 50% of our pro developers are using it, and it’s the fastest growing language on GitHub It’s production ready, and it’s being used by top apps both inside and outside of Google One of the concerns that we have, though, is that developers’ typing skills are eroding without the need for so much boilerplate code (cheers and applause) In fact, developers are so desperate for finger exercise that they’re actually starting to comment their Kotlin code (laughter) A little But you’ve asked us to do even more for Kotlin, so we’re announcing today a big step Android is becoming increasingly Kotlin-first, with many new Jetpack APIs being introduced first for Kotlin developers (cheers and applause) We also think that if you have a new project that you’re starting that you should be doing it in Kotlin, because, frankly, there’s less to type, there’s less to test, and there’s less to maintain We’re also investing much more in tooling, in docs, and in training For example, together with JetBrains, we’re announcing Kotlin/Everywhere, which is a global series of educational events But maybe you’re using one of Android’s two other officially-supported languages, C++ or the Java programming language That’s fine We continue to support these languages, and we continue to invest in them For example, we have a new toolchain for C++ On the Java side, we’re offering desugaring for language features up through Java 10, and soon you’ll see desugaring for popular OpenJDK libraries like Time and Streams So you can keep using those languages if they’re using them, and trust that we will keep supporting them We have always been committed on Android to compatibility, interoperability and our ecosystem,

and that is not changing Last year we announced Android Jetpack, which is a set of APIs to accelerate Android application development We’ve had great adoption so far, with 80% of the top 1,000 apps shipping with Jetpack modules Jetpack is all about typing less code to target a larger audience, because we support releases all the way back to API 14 It builds on the capabilities that we already had with a support library, adds in what we had introduced with the Architecture Components, and we just keep introducing new APIs and libraries as we go I’m going to talk about three new developments in Jetpack today– a new Camera library, some new Architecture Components, and a brief glimpse into something brand new we’re working on So today we’re launching CameraX It’s a library to make Camera application development much more easy and consistent across 90% of the Android devices out there A few years ago, I was talking to a couple of engineers from a company that was working on an interesting photo app It took great pictures, it did all these interesting filtering effects, but they had a problem Some of the devices the users had didn’t run their application correctly without workarounds in their code It was because of inconsistent support for some of the platform APIs So here’s how CameraX helps It works on releases all the way back to Lollipop, and it can take those workarounds and put it in the library itself So you call one API, and it figures how to make it work on the runtime device And it does that with more concise APIs for core use cases, so that your code can be smaller And, finally, it adds an extension add-on feature, which can access device-specific functionality like HDR, or night mode, or portrait mode We’re working with several manufacturers on the extensions right now, and you’ll soon see Extensions coming out for both new and existing devices Architecture Components came out a couple of years ago, specifically to address serious pain points in Android development It also works, like the rest of Jetpack, across versions all the way back to API level 14 We’ve had a couple of components that we introduced last year at I/O that recently went 1.0 One of those is WorkManager, for easier simplification of background job scheduling The other is Navigation Controller, for easier creation and editing of in-app navigation We’re also doing new work on existing components, such as RxJava and coroutine support in Room, as well as, for Kotlin developers, deep integration of coroutines into the Lifecycle and LiveData modules And we’re working on new components We announced SavedState for ViewModel, which allows easier processing of application or process restarts, and we also have a new Jetpack module for benchmarking which allows much easier performance testing of your application code And there’s one more thing I want to mention about Kotlin development We know that many of you have been wanting a modern reactive-style UI toolkit, one that takes advantage of Kotlin and also integrates with the platform code as well as all of your existing application code We’ve been working hard on a Kotlin-first library that we call Jetpack Compose It’s a reactive UI-programming library that’s We are proud to announce today we will be developing in the open in AOSP And that’s Sure, sure, you can do that (cheers and applause) So you can check that out starting today And this is just a brief tease If you want to know more about that, attend the “What’s New In Android” session which immediately follows this keynote So just stay in your seats and you’ll learn more Android Studio launched five years ago Since then they’ve had a lot of great features come out like ConstraintLayout editing, lots of profilers, static analysis, lots of great Kotlin support But we heard from you that some of the fundamentals didn’t work well enough for you, and it could be really frustrating So we had to revisit some of the basics The entire team, for the last six months, has stopped feature development completely to focus exclusively on quality, to make the core tool better (cheers and applause) They’ve fixed over 400 high-priority bugs– things like crashes and performance problems, UI freezes, memory leaks, and we’re always working on build speed as well We’ve also taken another run at some features to make them work better, such as Instant Run We heard from you that it was great when it worked, it wasn’t reliable enough, so we’ve rewritten that completely from the ground up with a much more stable foundation, and it’s now called Apply Changes The best part about all the quality work in Studio is it’s available today 3.5 Beta launches today, so you can go download it and play with it as soon as I’m done talking (cheers and applause) So having an application is great,

but getting it out to your users, even better So let’s talk about distribution We introduced Android App Bundles last year, and now there’s 80,000 applications shipping with this new bundling format And they’re getting an average APK size savings of about 20% Another thing that helps with size is Dynamic Feature Modules, which was in beta and is going 1.0 this week This allows you to choose the features that are downloaded onto a device, such as device-specific capabilities, or country information, or on-demand versus install-time features But both of these things are not just about size, they’re about the development and distribution process It’s much more modular, and it scales with team sizes much better So if you haven’t looked into them yet, we think that you should There’s one last area in distribution I want to talk about It’s in-app updates You’ve asked us for this feature for a while It’s the ability to have a really important bug that needs to be fixed– let’s say there’s a security problem, or an in-app billing that’s affecting monetization, or maybe you just have a really cool button that you want to put in the UI– and you want to make sure that your users get this update as soon as possible, without waiting for them to get to the Play Store and see that there’s an update available That capability is available now, and Tor is going to show you how it works (Tor Norbye) Thanks, Chet So, the in-app update functionality is now part of the Play Core library and is available for all of you to use right away But we’ve also started working on a Kotlin and Jetpack API to make it simpler to use, and that’s what I’m going to show you now So here I’m developing the Google I/O Conference app, and I’m going to open up my Activity And in the onCreate method I’m going to call a single extension method, updateIfRequired() and invoke Apply Changes In advance, I’ve already uploaded a higher version of my app to the Play Store to simulate a future update And, as you can see, now when users run my app they get this full-screen prompt asking them to update right away And note that I only had to add a single line of code to do that But this also means that every single update, even a minor one, will show this UI, and your users may not be really thrilled about that, So we can make the update check smarter We can pass in a lambda where we look at the offered version code and decide whether to show the full-screen prompt So that way, in the future, for each update you can choose whether to trigger the prompt by setting the right bits in the version code So now I’ve restarted my app, and this update did not request an immediate prompt so we’re back in our Activity And notice that there’s an Update button here now That’s something I added in advance This is our more flexible update API, and here’s the code for it I’m not going to explain this, because this part depends a lot on the surrounding app you’re integrating into But, as you can see, it’s not a lot of work or code So let’s take a look at what this lets you do, the flexible update API So this Update button only shows up if there’s an update available As a user, I can click on it to trigger an update That will then download in the background while I can continue to use the app And, when it’s done, I have the option of restarting when I am ready to Now, final important point You need to have the foresight to put an update check into your current version That’s what will let you push out an update to it later, when you might need it (Chet Haase) Thanks, Tor (applause) We had a really quick look today at how we’re making things better for both users and developers, as we all make Android a better product and platform– from privacy features, to Kotlin, to Jetpack, I really hope you enjoy the things that we’ve put in our… Q (laughter) Thank Q (laughter and applause) So that is my Q to leave, but first One of the exciting things in Android in the last couple of years is the collaboration that we’ve had with the Google Assistant Here to tell you more about the Assistant, here’s Chris Turkstra Thank you ♪ (music) ♪ (Chris Turkstra) Thank you, Chet We believe technology is at its best when it makes things simple, and that’s really what the Google Assistant is all about The idea that you can simply ask for what you want without thinking about all the steps involved is a fundamental shift in the way that people use technology And this isn’t just an idea People all around the world are using the Assistant every day to get things done– on their mobile devices, in their cars, and at home And in the smart home space alone, there are more than 30,000 unique devices

that now work with the Google Assistant These are built by a growing community of over 3,500 partners It’s momentum like this that creates lots of opportunities for developers like you Today, I’ll share new tools for content creators and app developers to build with the Assistant Let’s start with what’s new for content creators We know you want easier ways to reach more users across all of Google, and so we’re introducing a simple way to make your web content stand out across both Google Search and the Assistant, using something you’re already familiar with: markup We’re starting by enabling the support for the HowTo item type from so that your site’s how-to content can appear in a rich, structured way Here’s an example My kids are headed off to college soon, so we’re looking for a kid-replacement unit, also known as a dog (laughter) So I’ve been trying to figure out how to install a dog door Well, DIY Networks has an article on this, and they’ve implemented markup, identifying each step in their page So when I search for “How to Install a Dog Door” their content appears as a more structured, helpful result You’ll notice these step-by-step visual instructions on the search result page, and they really stand out And the best part is that the same simple markup creates an interactive result on its system-enabled smart displays This extends the reach of your content to an entirely new surface with no extra work In addition to instructional content on web pages, people also turn to YouTube every day to learn how to do things So we’re making it much easier for video creators to add content to the Assistant We’re adding a new How-To Template in the Actions console where you can turn your existing videos into interactive tutorials Let’s take a look Here, REI filled out a How-To Template with titles, text, and timestamps for each step in their video and uploaded it to the Actions console This transformed their video into an interactive, step-by-step experience with very little work You can get started with the HowTo Templates today Now let’s talk about what’s new for app developers, and I hear there might be a few Android fans here in the audience (applause) Thank you And we know you want to make it easier for users to get into your apps So last year we previewed App Actions, a way to create voice-based entry points from the Assistant to exactly the right spot in your app using intents Today we’re announcing four categories of App Actions that are ready to use– Health & Fitness, Finance, Ridesharing, and Food Ordering Let’s look at Health & Fitness, an area that I’m clearly new to (laughter) Nike has a great app that lets me track my runs, but when my shoes are on, my headphones are in, and I’m finally ready to actually take a run, I don’t want to tap through my phone to get it all started Wouldn’t it be much faster if I could just use my voice? Well, luckily, Nike implemented App Actions, so when I say, “Hey Google, start my run with Nike Run Club,” the Assistant fast forwards into Nike’s app and automatically starts my run I didn’t need to swipe, tap, or navigate to find what I needed, I simply asked Let’s take a look at how they did it Here you can see Nike’s actions.xml file in their manifest where they map the START_EXERCISE built-in intent to the part of their app that starts a workout And with that change, their redeploy, their APK, and it’s ready for action You can get started today with the four live intent categories And we have more for app developers Smart displays have been a huge hit for consumers And, for the first time ever, we’re opening up the ability for app developers to address the full display of the smart display So we’re introducing a developer preview of Interactive Canvas, a tool to create full-screen experiences for smart displays that leverage voice, visuals, and touch It uses open web technologies you’re already familiar with, like HTML, CSS, and JavaScript And we’re starting where interactive experiences really come to life: games Here’s an example of a trivia game– it was built by HQ University with Interactive Canvas– that leverages the full-screen experience You can start building games today,

and we’d love to hear your feedback as we consider which categories to open up next Whether you’re looking to reach more users across Google with your content, drive more engagement with your apps, or build custom experiences for the Assistant, you now have more tools to do so And now I’d like to introduce Tal, who’s going to talk more about the web platform Thank you (applause) ♪ (music) ♪ (Tal Oppenheimer) Thanks, Chris The open web provides universal access to the world’s information and services for billions of people across the full spectrum of devices– from entry-level phones, to high-powered desktops With Chrome, we’re focused on providing the global community with a modern, continuously updated browser with new releases every six weeks And we contribute our work to the entire web ecosystem with the open source Chromium project Today we’re going to share the latest improvements to Chrome and our developer tools that you can use to make your websites faster and more powerful, all while keeping user trust front and center We all know that speed matters on the web, and that even small improvements can translate into big wins for developers and for users So we’re continuously working to make the web faster One area that we focused on with Chrome is reducing that startup time, so that now, when you launch Chrome on an entry-level Android phone, the web page loads in almost half the time And this speed-up is partly due to improvements in V8, our open source JavaScript engine It’s now two times faster at parsing JavaScript, and uses up to 20% less memory on real-world websites In addition to these improvements in Chrome, we’re also adding features to the web platform to help you make your sites even faster As one example, there’s image lazy loading Modern websites are more visual than ever, using lots of beautiful high-resolution imagery But loading all those images at once can bog down the browser and can waste the user’s data by loading unnecessary images that the user never actually sees So it’s often better to load images only as they’re actually needed, a technique known as lazy loading We know it can be a lot of work for developers to use their own JavaScript solutions, and it can be hard to get the quality experience you want for your users So we wanted to make it incredibly simple to have a great image-loading experience on your site Starting behind a flag in Chrome Canary today, all you’ll need to do is add the new loading attribute to your image tags, and Chrome will take care of the rest (cheers and applause) We’ll take into account factors like connection speed to decide when to load the images, and we’ll check the first two kilobytes of the deferred image to add placeholders that are the right size The end result is a much smoother experience for image-heavy sites, all without the need to write any extra code We’ve also been enhancing our developer tools to help you understand how you can use all of these improvements Three years ago, we introduced Lighthouse, a powerful tool that audits your website and provides you with clear guidance on how you can improve your site’s performance, security, and lots more Lighthouse’s reports have been used to improve millions of web pages, but because our websites, and the web itself, are constantly evolving, auditing tools like Lighthouse need to become a continual part of the development process And we’re seeing that some developers are already doing this At Pinterest, they’ve set specific size limits and performance thresholds for their site, and continuously measure it to ensure they aren’t regressing They call these limits a “budget,” and by optimizing for performance and enforcing these budgets they’re able to ensure that their site remains fast and delivers great results And we think this is a fantastic practice, so we’ve added support for performance budgets directly to Lighthouse Now you can set budgets based on size, such as total JavaScript download, or for target metrics like page load time And by integrating Lighthouse with your continuous integration server you can ensure that your site stays svelte and healthy In addition to making the web faster, we’re also working to make the web more powerful and more deeply integrated with devices and operating systems And Google Duo is a great example of this It uses the latest features of the web to support high-quality video calling right in the browser

And using WebAssembly, the team was able to easily bring their native app features, like echo detection, to the web And because Duo for Web is a progressive web app, users can install it across Chrome OS, macOS, Linux, and Windows, so it’s launchable from a desktop icon and it runs in its own window And we’re also seeing companies like Hulu taking advantage of these new capabilities to deliver a powerful, immersive desktop-app experience and driving more repeat usage And Twitter shows the power of these progressive web apps with a single codebase that scales seamlessly across a wide variety of devices And Twitter and Hulu will actually be joining us onstage later today at I/O, to tell us more about their experiences As the web and browsers continue to evolve, we know that search engines need to keep up in order to understand how to properly index modern websites with these latest features Today we’re announcing that Google Search is now using the latest version of Chromium to index the web This means that, as a developer, you can focus on building your site using the latest web platform features, without having to worry about using hacks or workarounds to ensure that the Google crawler properly sees your content And you can learn more on our Webmaster blog, or at our session here at I/O on Thursday User trust and safety is at the heart of everything we do in Chrome It’s motivated features like Safe Browsing, that protects users from phishing attacks and malware sites, and our recent efforts to move all web traffic to HTTPS We believe that giving users transparency, choice, and control over how their data is used and shared on the web is an important part of these efforts, and it requires that we all rethink some of the fundamentals of the web Today we’re sharing three changes that we’ll start rolling out later this year First, we’re changing how cookies work in Chrome, making them more private and secure by default Second, we’re adding new features in Chrome that will build on these cookie changes to give users more transparency and easy-to-use controls over how sites track them across the web Third, we’re enhancing Chrome to protect users from techniques like fingerprinting that are used to bypass this user choice and control As we work to make the web safer, we’re committed to preserving the health of the overall web ecosystem and to working in cooperation with the broader developer community You can learn more about these upcoming changes and what they mean for you on our Chromium blog and in the “What’s New with Chrome on the Web” session here at I/O As the web continues to evolve at a rapid pace, we know it can be a little bit tricky to keep up with the latest features and best practices We created a new website called to help It’s a simple, straightforward guide to teach you how to build on the modern web with interactive code labs on the most important topics including how to optimize popular web frameworks like REACT for top performance can help you use the best of the modern web platform to create fast, powerful experiences for your users In addition to these platform and tool changes, we’ve also been improving Chrome OS In Q4, Chrome OS accounted for 21% of U.S. notebook sales, and we expect to see growth continue as more devices hit the market Last year we brought Linux support to Chromebooks, and we’ve continued to integrate this to provide you with an easy-to-use Linux development environment that’s fast and safe, thanks to Google’s best-in-class VM and Sandbox isolation technology This integration allows you to do things like seamlessly share files across Chrome OS, Google Drive, Android, and Linux, and you get all the standard Linux features, like port forwarding, that lets you run your Web server in the Linux container and debug on the same machine All of this makes Chrome OS a great choice for developers Web developers can code in a familiar Linux environment and can test on a variety of desktop and mobile browsers And Android Studio is now available with a one-click install and integrated debugging on any of the Android developer recommended Chromebooks And, finally, we’re happy to announce that all Chromebooks launched this year will be Linux-ready right out of the box (cheers and applause) And these are just a few of the most recent improvements across the web and Chrome OS Stay tuned for more features coming later this year Now I’d like to invite Anitha

to talk to you about machine learning and AI (cheers and applause) ♪ (music) ♪ (Anitha Vijayakumar) Thanks, Tal Here at Google, we are using AI to solve a range of challenging problems to help people in their daily lives Whether it is building an AI system to get rid of 100-million spam messages each day in Gmail, or serving relevant and up-to-date content through Google News using new AI techniques, or training a neural network to improve the way we interact with our mobile devices through faster speech recognition on Gboard, AI enables us to make our products more helpful and delightful for our users But we want all developers to be able to build successful AI-enabled applications to solve problems That is why we built AI tools that are easy to use right out of the box, powerful, and flexible, all available on the platforms that you care about Our tools span from APIs that are easy to get started with and come ready out of the box, like ML Kit, to more powerful tooling that gives you the performance you need like AutoML and cloud TPUs in the Google Cloud, and the flexibility to run powerful AI systems in production anywhere with open source libraries like TensorFlow For the millions of Android and iOS developers, ML Kits get you started with common machine learning tasks immediately It’s taking some of the best features that power Google’s own AI applications and making them available to you Since launching it at I/O last year, we’ve seen a lot of adoption in a wide array of use cases, like the TextPlus app which uses SmartReply in-app to suggest responses, or the Gradeup app using text recognition to help students scan in their homework ML Kit’s momentum is why we are excited to share a couple of new announcements First, we are making AI-powered translation available through a new on-device translation API This new API provides fast, dynamic translation for 59 languages Now, the same machine learning models that power the Google’s Translate app are available for your app even when they’re offline, helping to save energy and reduce latency Second, ML Kit can help power your app’s visual search experience with a new object detection and tracking API For retail apps, you can pair this with Google Cloud’s product Search API to match your product SKUs Our partners are already using this API to build helpful AI experiences for their users, like this beautiful IKEA app which lets you search for products by just pointing your mobile phone camera These are just a few examples of how ML Kit makes it easy to integrate AI into your Android and iOS apps You can get started using these new APIs today through Firebase For developers who need more performance for large machine learning workloads, Google Cloud provides a complete set of AI tools, and this includes AutoML AutoML allows you to train accurate models on your own datasets without having to write a single line of code And we recently expanded this product family to support new use cases, like the new AutoML Tables where you can ingest structured tabular datasets and generate prediction models in days instead of weeks Now you can take your large CSV files or database tables and use them to tackle a range of challenging problems such as fraud detection, optimizing lead conversion, and predicting user demand What’s amazing here is that you can do all of this with a few clicks, no coding required, and generate a world-class machine learning model We also launched AutoML Video Intelligence This new service lets you create custom models that automatically classify your video content with the labels that you define So if you deal with a large dataset and you want to analyze your content by what is in each frame, you can create custom labels and AutoML Video Intelligence will help you categorize the content and make it easily searchable But to train and deploy your models, you need powerful computational resources like cloud TPUs, our custom AI hardware accelerators Through Google Cloud, our partners have been using Cloud TPU devices to accomplish large-scale machine learning tasks

at incredible speeds For example, Recursion Pharmaceuticals is using Cloud TPUs to analyze cellular microscopic images to help treat rare diseases By using cloud TPUs, they’re able to reduce their training time from 24 hours to nearly 15 minutes Today we’re announcing that you can use Cloud TPU V3 pods in beta on the Google Cloud These pods are made up of individual TPUs and, when assembled together, you can train models for image classification, natural language processing, and many other ML applications, scaling up by adding just two lines of code Cloud AutoML and Cloud TPU Pods are meant to provide you with the performance you need to speed up your machine learning workflow We also want to keep you in complete control and give you flexibility That is why we decided to open source TensorFlow TensorFlow is helping democratize AI among developers, businesses, and researchers by helping them build truly customized AI experiences Since open sourcing it in 2015, it has matured into a flexible end-to-end machine learning ecosystem with a global community We recently announced TensorFlow 2.0 in alpha, with plans to launch a release candidate soon TensorFlow 2.0 is all about usability We are making it even easier to build and deploy custom models with more intuitive APIs, less code needed, and more flexible for powerful experimentation and deployment on the platforms that you care about For JavaScript developers, TensorFlow.js helps you build, train, and deploy custom models right in the browser and on the Node.js platform For developers working with on-device platforms such as mobile phones and IoT, TensorFlow Lite can help you address common obstacles in your AI-powered apps like poor network connectivity, protecting user privacy, and low-latency environments, without sacrificing performance In just 18 months, TensorFlow Lite has been installed on more than 2 billion devices including Android, iOS, and embedded systems To demonstrate how fast and flexible TensorFlow Lite is, Tim is going to show you custom models, tracking user movement in real-time using the GPU on-device (Tim Davis) Thanks, Anitha We’ve built a super fun app called Dance Like which helps anyone learn how to be a better dancer using machine learning I’ll demo it in a bit, but before I do, I want to say that making this app was a lot of hard work and was only made possible because of a bunch of teams at Google but most importantly because of TensorFlow Lite So why was it so tricky? Well, we set ourselves the goal of running five intensive, on-device tasks in parallel, in real-time, without sacrificing performance These tasks were: running two body-part segmentation models, matching the segmentation models, running dynamic time warping, playing a video, and encoding a video And let me emphasize this again– all on-device, simultaneously, in real-time To accomplish this, TensorFlow Lite enables us to easily delegate acceleration of our ML models on a GPU You can do this on both iOS and Android, so you have a single framework for your on-device ML Here’s a code snippet which shows how we set up this delegation I created a TensorFlow Lite interpreter And to execute my model on the GPU, I just construct a new GPU delegate and modify the interpreter to use the GPU delegate, and I’m done The models will now run on the GPU Now I’m going to do a demo live and walk you through the five on-device tasks running at the same time Here we go. Wish me luck! So there’s a few dances you can choose from I’m going to start with slow motion, because I’m a beginner Now, as I fire up my dance moves, you can see the real-time segmentation model running on me It’s segmenting me out from the background and identifying all the different parts of my body Now, as I follow along the dancer, a second segmentation model is now running

So now there’s two segmentation models running, via the GPU, to produce a matching score The matching score, up in the top right-hand corner, is telling me how well I’m matching the dancer’s moves How awesome is that, right? (cheers and applause) But, wait, there’s more! You know what is really cool? Using dynamic time warping! This syncs my slow-motion moves with my real-time dance Pretty fun, right? You can tell this was just another day at Google Back to you, Anitha (Anitha) Isn’t that cool? Nice moves, Tim! (applause) Whether you are just getting started with machine learning or building custom AI apps like Dance Like, we want to help every developer build incredible AI applications, whatever challenging problems you are trying to solve Now, to talk more about what we are doing in mobile development, I’d like to introduce Kristen Johnson Thank you (applause) ♪ (music) ♪ Thanks, Anitha It’s great to be here with all of you to talk a bit about Firebase Our mission is to help mobile and Web app developers, just like all of you, be successful We provide you with a platform of tools and Cloud services that simplify your app dev workflows and infrastructure needs so you can focus on building amazing user experiences With Firebase, you can build your app with fully managed back-ends, improve your app’s quality with testing and monitoring, and engage your users with better insights As you heard earlier, Google is committed to making AI available to every developer Firebase and ML Kit make it easy for you to bring machine learning to your apps, regardless of expertise And today we’re expanding ML Kit with the addition of AutoML Vision Edge This will simplify the workflow of building and training your custom TensorFlow Lite models to classify images All you have to do is upload a set of images, click a button to train your model, and then publish it That’s it Now, once published, your model is hosted on Google’s infrastructure, and, with just a few lines of code, your app can dynamically retrieve the model and run it on-device I’d like to invite my teammate Stella on stage to show us how it works (Stella Gaitani) Thanks, Kristen For this demo, I want to build an app that identifies different dog breeds I will start in the Firebase console in the ML Kit section by creating a new dataset which I will call dog_breeds Once my dataset is created, I just need to add my images I will grab my data from my machine, and I’m going to upload it here Now, this can take a little while, so to speed things up, I will jump over to this dataset where I have already uploaded the images As you can see, here I can view all the images I uploaded, organized by label And I can easily add or remove images The next step is to train the model When training the model, I can choose how large my model should be to optimize between latency and accuracy And I can also choose how long to train my model for Typically, larger datasets require longer training times For this demo, I will choose eight hours because my dataset is quite big Again, to speed things up, I’m going to jump over to this dataset where the model has already been trained Once the model finishes training, Firebase will provide me with an evaluation to help me decide if this model meets my needs or if I should continue iterating I can see the precision and recall rates And below these, I have a full breakdown of how often my model labeled an image correctly for each label The next and final step is to publish the model Once the model is published, it will be hosted on Google’s infrastructure What this means is that I can add a few lines of code to my app, and then my app will be able to dynamically download the model and do on-device inferencing And, now, let’s see if this works! Unfortunately, I couldn’t bring my dog with me on stage, so we will have to use a stand-in Here is our friend (laughter)

And I’m going to launch the app and I’m going to aim a camera at our friend here It’s a little harder with a stuffed animal, but here we go You can see that my app in some angles– This is a border collie, so it can pick up the border collie right here And all of this was done with just a few lines of code and a few button clicks (applause) Thanks, Stella So that’s just one way that we’re expanding Firebase to simplify app development for you Another area that we aim to simplify is optimizing the performance of your apps Firebase Performance Monitoring gives you insights into the start-up time and network responsiveness of your iOS and Android apps Now we’ve heard from all you web developers out there that you want the same real user monitoring that native developers get Well, today we’re expanding Performance Monitoring to the web! (applause) You heard from Tal that users want the web to be fast, which means performance is a major driver for success on the web Tools like Lighthouse are great for giving you a picture of how your web app performs in a targeted, synthetic test Now, with Firebase, you can complement these tools with an understanding of how your users are experiencing your web app out in the wild This is awesome, because it’s the first time that you can see metrics that are more granular than just page load time, like how long it takes users to see any content on your page or if the page is ready for interaction And you see the full distribution curve of these metrics across different countries, browsers, and network connections As an example, I might notice that load times are slow for users in a region When I investigate further, I uncover that there is an issue with a particular CDN point of presence Without performance monitoring, I would not have understood why my usage was declining in that region And all of these insights come from just a few lines of code Firebase Performance Monitoring for the web is available for free in beta starting today (applause) And these are just two of the many exciting updates we’re launching here at I/O for Firebase, from AutoML Vision Edge to our new Performance Monitoring for web apps With every improvement to Firebase, we aim to simplify your app dev workflows and infrastructure needs so that you can stay focused on building amazing user experiences And now I’m going to turn it over to Adam to wrap things up (applause) ♪ (music) ♪ Thank you, Kristen So I lead Developer Relations, which gives me a really unique vantage point It’s great to connect with developers every day We get to see how all of you are making web apps, mobile apps, new AI-driven experiences, and we get to hear the feedback that helps us make our products better And I’m excited, because the team has built a really great I/O for you this year We’ve got sandboxes where you can see our products in action, and you’ll find everything from AR, the new serverless cloud offerings we’ve released, and actually an entire sandbox dedicated to our AI and ML offerings Across I/O you’ll find more than 180 technical sessions this year, all presented by engineering and product leads We’ve added a new gaming track that I think you’ll really like with everything from Android, Cloud gaming, to Stadia, which is pretty exciting And last year, you told us you liked the inspiration sessions, so I’m excited to see some of what we’re bringing We’re bringing even more this year, and I’m excited to hear from astronaut Mae Jemison, Turing Award winner Geoffrey Hinton, and actually Wayne Coyne, the lead singer of the Flaming Lips, is doing an inspiration session Who, by the way, The Flaming Lips are also going to put on an incredible show tomorrow night I’m incredibly excited to see it We have a cool AI musical integration that they’ve done And for everybody watching online, we’re broadcasting it live so you can watch it too (applause) Now, we have a ton of things to show you at I/O, a ton of new technology that Google is releasing, and one of my favorites I wanted to take a minute on is Flutter (cheers and applause) People love Flutter because it’s our open source toolkit for building iOS and Android mobile apps from a single codebase But we’re really pleased to announce something new today,

and that is a technical preview of Flutter for the Web (cheers and applause) So now you can take the same code you would use for mobile devices and bring it to the web So an example of this is The New York Times, and The New York Times are doing this with their puzzle apps And rather than rewriting the puzzle app for every different platform, they can just write it once with Flutter And they can even use Flutter to take capabilities like this and add it to their existing apps and include Flutter technology in it So here’s an example This is KenKen, one of their number puzzles And the game changer here is they’ve got the same app with the same codebase running at 60 frames per second on Android, on iOS, on Mac, on Windows, and now on the web It’s all compiled automatically, down in native code or JavaScript It’s pretty cool stuff This is one of the things we’ll be showing We’ll have this in the Flutter Sandbox Check it out today (cheers and applause) So now I wanted to take a second and talk about the Google developer community All of you are here, we’re ten years ten years of our Google developer community with programs like Google Developer Groups, Google Developer Experts, LaunchPad, Women Techmakers (cheers) You form connections with developers all over the world, and I wanted to take a minute to tell you a story about one of those developers I wanted to tell you a story about a woman named Nazirini Nazirini is a developer from Uganda, and in that country they’re dealing with a pest in their crops called fall armyworm, and it’s devastating Nazirini is a developer who got introduced to TensorFlow through a Study Jam, and she worked with a team and used her Android and her new TensorFlow skills to build an app that uses machine learning to help diagnose and detect this attack of fall armyworm earlier to help the farmers treat it and save their harvest Nazirini is actually here today, and I thought we could just give her a big hand (cheers and applause) So, as you can see, these connections you make with the community really matter, and make a difference in the world Thank you Thank you for being part of this community For the meet-ups you host, for all the feedback you give us, and all the giving back and mentoring you do to other developers, thank you And if anyone here is not already part of one of those communities, there’s absolutely no reason why you shouldn’t join And you can join online with your local group or you can actually do it here We built a really cool developer Community Lounge just behind us here that I think you’ll really enjoy hanging out with So then, finally, it’s also about all the connections we have here at Google with you And I’m excited that we have more than 2,500 Googlers here to meet with you, to talk with you, to listen to you, to get your feedback and share ideas, and we’re incredibly excited to hear how you’re using our platforms So with that, I/O is off, it’s off to a great start The sessions are about to begin Thanks, everyone for coming to I/O Have a great time (cheers and applause) ♪ (music) ♪