TensorFlow World 2019 Keynote

JEFF DEAN: I’m really excited to be here I think it was almost four years ago to the day that we were about 20 people sitting in a small conference room in one of the Google buildings We’ve woken up early because we wanted to kind of time this for an early East Coast launch where we were turning on the TensorFlow.org website and releasing the first version of TensorFlow as an open source project And I’m really, really excited to see what it’s become It’s just remarkable to see the growth and all the different kinds of ways in which people have used this system for all kinds of interesting things around the world So one thing that’s interesting is the growth in the use of TensorFlow also kind of mirrors the growth in interest in machine learning and machine learning research generally around the world So this is a graph showing the number of machine learning archive papers that have been posted over the last 10 years or so And you can see it’s growing quite, quite rapidly, much more quickly than you might expect And that lower red line is kind of the nice doubling every couple of years growth rate, exponential growth rate we got used to in computing power, due to Moore’s law for so many years That’s now kind of slowed down But you can see that the machine learning research community is generating research ideas at faster than that rate, which is pretty remarkable We’ve replaced computational growth with growth of ideas, and we’ll see those both together will be important And really, the excitement about machine learning is because we can now do things we couldn’t do before, right? As little as five or six years ago, computers really couldn’t see that well And starting in about 2012, 2013, we started to have people use deep neural networks to try to tackle computer vision problems, image classification, object detection, things like that And so now, using deep learning and deep neural networks, you can feed in the raw pixels of an image and fairly reliably get a prediction of what kind of object is in that image Feed in the pixels there Red, green, and blue values in a bunch of different coordinates, and you get out the prediction leopard This works for speech as well You can feed an audio wave forms, and by training on lots of audio wave forms and transcripts of what’s being said in those wave forms, we can actually take a completely new recording and tell you what is being said amid a transcript Bonjour, comment allez-vous? You can even combine these ideas and have models that take in pixels, and instead of just predicting classifications of what are in the object, it can actually write a short sentence, a short caption, that a human might write about the image– a cheetah lying on top of a car That’s one of my vacation photos, which was kind of cool And so just to show the progress in computer vision, in 2011, Stanford hosts an ImageNet contest every year to see how well computer vision systems can predict one of 1,000 categories in a full color image And you get about a million images to train on, and then you get a bunch of test images your model has never seen before And you need to make a prediction In 2011, the winning entrant got 26% error, right? So you can kind of make out what that is But it’s pretty hard to tell We know from human experiment that human error of a well-trained human, someone who’s practiced at this particular task and really understands 1,000 categories, gets about 5% error So this is not a trivial task And in 2016, the winning entrant got 3% error So just look at that tremendous progress in the ability of computers to resolve and understand computer imagery and have computer vision that actually works This is remarkably important in the world, because now we have systems that can perceive the world around us and we can do all kinds of really interesting things about We’ve seen similar progress in speech recognition and language translation and things like that So for the rest of the talk, I’d like to kind of structure it around this nice list of 14 challenges that the US National Academy of Engineering put out and felt like these were important things for the science and engineering communities to work on for the next 100 years They put this out in 2008 and came up with this list of 14 things after some deliberation And I think you’ll agree that these are sort of pretty good large challenging problems, that if we actually make progress on them, that we’ll actually have a lot of progress in the world We’ll be healthier We’ll be able to learn things better We’ll be able to develop better medicines We’ll have all kinds of interesting energy solutions So I’m going to talk about a few of these And the first one I’ll talk about is restoring and improving urban infrastructure So we’re on the cusp of the sort of widespread commercialization

of a really interesting new technology that’s going to really change how we think about transportation And that is autonomous vehicles And this is a problem that has been worked on for quite a while, but it’s now starting to look like it’s actually completely possible and commercially viable to produce these things And a lot of the reason is that we now have computer vision and machine learning techniques that can take in sort of raw forms of data that the sensors on these cars collect So they have the spinning LIDARs on the top that give them 3D point cloud data They have cameras in lots of different directions They have radar in the front bumper and the rear bumper And they can really take all this raw information in, and with a deep neural network, fuse it all together to build a high level understanding of what is going on around the car Or is it another car to my side, there’s a pedestrian up here to the left, there’s a light post over there I don’t really need to worry about that moving And really help to understand the environment in which they’re operating and then what actions can they take in the world that are both legal, safe, obey all the traffic laws, and get them from A to B And this is not some distant far-off dream Alphabet’s Waymo subsidiary has actually been running tests in Phoenix, Arizona Normally when they run tests, they have a safety driver in the front seat, ready to take over if the car does something kind of unexpected But for the last year or so, they’ve been running tests in Phoenix with real passengers in the backseat and no safety drivers in the front seat, running around suburban Phoenix So suburban Phoenix is a slightly easier training ground than, say, downtown Manhattan or San Francisco But it’s still something that is like not really far off It’s something that’s actually happening And this is really possible because of things like machine learning and the use of TensorFlow in these systems Another one that I’m really, really excited about is advance health informatics This is a really broad area, and I think there’s lots and lots of ways that machine learning and the use of health data can be used to make better health care decisions for people So I’ll talk about one of them And really, I think the potential here is that we can use machine learning to bring the wisdom of experts through a machine learning model anywhere in the world And that’s really a huge, huge opportunity So let’s look at this through one problem we’ve been working on for a while, which is diabetic retinopathy So diabetic retinopathy is the fastest growing cause of preventable blindness in the world And screening every year, if you’re at risk for this, and if you have diabetes or early sort of symptoms that make it likely you might develop diabetes, you should really get screened every year So there’s 400 million people around the world that should be screened every year But the screening is really specialized Doctors can’t do it You really need ophthalmologist level of training in order to do this effectively And the impact of the shortage is significant So in India, for example, there’s a shortage of 127,000 eye doctors to do this sort of screening And as a result, 45% of patients who are diagnosed to this disease actually have suffered either full or partial vision loss before they’re actually diagnosed and then treated And this is completely tragic because this disease, if you catch it in time, is completely treatable There’s a very simple 99% effective treatment that we just need to make sure that the right people get treated at the right time So what can you do? So, it turns out diabetic retinopathy screening is also a computer vision problem, and the progress we’ve made on general computer vision problems where you want to take a picture and tell if that’s a leopard or an aircraft carrier or a car actually also works for diabetic retinopathy So you can take a retinal image, which is what the screening camera, sort of the raw data that comes off the screening camera, and try to feed that into a model that predicts 1, 2, 3, 4, or 5 That’s how these things are graded, 1 being no diabetic retinopathy, 5 being proliferative, and the other numbers being in between So it turns out you can get a collection of data of retinal images and have ophthalmologists label them Turns out if you ask two ophthalmologists to label the same image, they agree with each other 60% of the time on the number 1, 2, 3, 4, or 5 But perhaps slightly scarier if you ask the same ophthalmologist to grade the same image a few hours apart, they agree with themselves 65% of the time But you can fix this by actually getting each image labeled by a lot of ophthalmologists, so you’ll get it labeled by seven ophthalmologists If five of them say it’s a 2, and two of them say it’s a 3, it’s probably more like a 2 than a 3 Eventually, you have a nice, high quality data set you can train on Like many machine learning problems, high quality data is the right raw ingredient

But then you can apply, basically, an off-the-shelf computer vision model trained on this data set And now you can get a model that is on par or perhaps slightly better than the average board certified ophthalmologist in the US, which is pretty amazing It turns out you can actually do better than that And if you get the data labeled by retinal specialists, people who have more training in retinal disease and change the protocol by which you label things, you get three retinal specialists to look at an image, discuss it amongst themselves, and come up with what’s called a sort of coordinated assessment and one number Then you can train a model and now be on par with retinal specialists, which is kind of the gold standard of care in this area And that’s something you can now take and distribute widely around the world So one issue particularly with health care kinds of problems is you want explainable models You want to be able to explain to a clinician why is this person, why do we think this person has moderate diabetic retinopathy So you can take a retinal image like this, and one of the things that really helps is if you can show in the model’s assessment why this is a 2 and not a 3 And by highlighting parts of the input data, you can actually make this more understandable for clinicians and enable them to really sort of get behind the assessment that the model is making And we’ve seen this in other areas as well There’s been a lot of work on explainability, so I think the notion that deep neural networks are sort of complete black boxes is a bit overdone There’s actually a bunch of good techniques that are being developed and more all the time that will improve this So a bunch of advances depend on being able to understand text And we’ve had a lot of really good improvements in the last few years on language understanding So this is a bit of a story of research and how research builds on other research So in 2017, a collection of Google researchers and interns came up with a new kind of model for text called the Transformer model So unlike recurrent models where you have kind of a sequential process where you absorb one word or one token at a time and update some internal state and then go on to the next token, the Transformer model enables you to process a whole bunch of text, all at once in parallel, making it much more computationally efficient, and then to use attention on previous texts to really focus on if I’m trying to predict what the next word is, what are other parts of the context to the left that are relevant to predicting that? So that paper was quite successful and showed really good results on language translation tasks with a lot less compute So the blue score there and the first two columns for English to German and English to French, higher is better And then the compute cost of these models shows that this is getting sort of state of the art results at that time, with 10 to 100x less compute than other approaches Then in 2018, another team of Google researchers built on the idea of Transformers So everything you see there in a blue oval is a Transformer module, and they came up with this approach called Bidirectional Encoding Representations from Transformers, or BERT It’s a little bit shorter and more catchy So BERT has this really nice property that, in addition to using context to the left, it uses context all around the language, sort of the surrounding text, in order to make predictions about text And the way it works is you start with a self-supervised objective So the one really nice thing about this is there’s lots and lots of text in the world So if you can figure out a way to use that text to train a model to be able to understand text better, that would be great So we’re going to take this text, and in the BERT training objective, to make it self-supervised, we’re going to drop about 15% of the words And this is actually pretty hard, but the model is then going to try to fill in the blanks, essentially Try to predict what are the missing words that were dropped And because we actually have the original words, we now know if the model is correct in its guesses about what goes in the box And by processing trillions of words of text like this, you actually get a very good understanding of contextual cues in language and how to actually fill in the blanks in a really intelligent way And so that’s essentially the training objective for BERT You take text, you drop 15% of it, and then you try to predict those missing words And one key thing that works really well is that step one You can pre-train a model on lots and lots of text, using this fill-in-the-blank self-supervised objective function And then step two, you can then take a language task you really care about Like maybe you want to predict, is this a five-star review

or a one-star review for some hotel, but you don’t have very much labeled text for that actual task You might have 10,000 reviews and know the star count of each review But you can then finetune the model, starting with the model trained in step one on trillions of words of text and now use your paltry 10,000 examples for the text task you really care about And that works extremely well So in particular, BERT gave state-of-the-art results across a broad range of different text understanding benchmarks in this GLUE benchmark suite, which was pretty cool And people have been using BERT now in this way to improve all kinds of different things all across the language understanding and NLP space So one of the grand challenges was engineer the tools of scientific discovery And I think it’s pretty clear machine learning is actually going to be an important component of making advances in a lot of these other grand challenge areas, things like autonomous vehicles or other kinds of things And it’s been really satisfying to see what we’d hoped would happen when we released TensorFlow as an open source project has actually kind of come to pass, as we were hoping, in that lots of people would sort of pick up TensorFlow, use it for all kinds of things People would improve the core system They would use it for tasks we would never imagine And that’s been quite satisfying So people have done all kinds of things Some of these are uses inside of Google Some are outside in academic institutions Some are scientists working on conserving whales or understanding ancient scripts, many kinds of things, which is pretty neat The breadth of uses is really amazing These are the 20 winners of the Google.org AI Impact Challenge, where people could submit proposals for how they might use machine learning and AI to really tackle a local challenge they saw in their communities And they have all kinds of things, ranging from trying to predict better ambulance dispatching to identifying sort of illegal logging using speech recognition or audio processing Pretty neat And many of them are using TensorFlow So one of the things we’re pretty excited about is AutoML, which is this idea of automating some of the process by which machine learning experts sit down and sort of make decisions to solve machine learning problems So currently, you have a machine learning expert sit down, they take data, they have computation They run a bunch of experiments They kind of stir it all together And eventually, you get a solution to a problem you actually care about One of the things we’d like to be able to do, though, is see if we could eliminate a lot of the need for the human machine learning expert to run these experiments and instead, automate the experimental process by which a machine learning expert comes by a high quality solution for a problem you care about So lots and lots of organizations around the world have machine learning problems, but many, many of them don’t even realize they have a machine learning problem, let alone have people in their organization that can tackle the problem So one of the earliest pieces of work our researchers did in the space was something called neural architecture search So when you sit down and design a neural network to tackle a particular task, you make a lot of decisions about shapes of this or that, and should it be used 3 by 3 filters at layer 17 or 5 by 5, all kinds of things like this It turns out you can automate this process by having a model generating model and train the model generating model based on feedback about how well the models that it generates work on the problem you care about So the way this will work, we’re going to generate a bunch of models Those are just descriptions of different neural network architectures We’re going to train each of those for a few hours, and then we’re going to see how well they work And then use the accuracy of those models as a reinforcement learning signal for the model generating model, to steer it away from models that didn’t work very well and towards models that worked better And we’re going to repeat many, many times And over time, we’re going to get better and better by steering the search to the parts of the space of models that worked well And so it comes up with models that look a little strange, admittedly A human probably would not sit down and wire up a sort of machine learning, computer vision model exactly that way But they’re pretty effective So if you look at this graph, this shows kind of the best human machine learning experts, computer vision experts, machine learning researchers in the world, producing a whole bunch of different kinds of models in the last four or five years, things like ResNet 50, DenseNet-201, Inception-ResNet, all kinds of things That black dotted line is kind of the frontier of human machine learning expert model quality on the y-axis and computational cost

on the x-axis So what you see is as you go out the x-axis, you tend to get more accuracy because you’re applying more computational cost But what you see is the blue dotted line is AutoML-based solutions, systems where we’ve done this automated experimentation instead of pre-designing any particular architecture And you see that it’s better both at the high end, where you care about the most accurate model you can get, regardless of computational cost, but it’s also accurate at the low end, where you care about a really lightweight model that might run on a phone or something like that And in 2019, we’ve actually been able to improve that significantly This is a set of models called Efficient Net and it has a very kind of a slider about you can trade off computational cost and accuracy But they’re all way better than human sort of guided experimentation on the black dotted line there And this is true for image recognition, for [INAUDIBLE] It’s true for object detection So the red line there is AutoML The other things are not It’s true for language translation So the black line there is various kinds of Transformers The red line is we gave the basic components of Transformers to an AutoML system and allowed it to fiddle with it and come up with something better It’s true for computer vision models used in autonomous vehicles So this was a collaboration between Waymo and Google Research We were able to come up with models that were significantly lower latency for the same quality, or they could trade it off and get significantly lower error rate at the same latency It actually works for tabular data So if you have lots of customer records, and you want to predict which customers are going to be spending $1,000 with your business next month, you can use AutoML to come up with a high quality model for that kind of problem OK So what do we want? I think we want the following properties in a machine learning model So one is we tend to train separate models for each different problem we care about And I think this is a bit misguided Like, really, we want one model that does a lot of things so that it can build on the knowledge in how it does thousands or millions of different things, so that when the million and first thing comes along, it can actually use its expertise from all the other things it knows how to do to know how to get into a good state for the new problem with relatively little data and relatively little computational cost So these are some nice properties I have kind of a cartoon diagram of something I think might make sense So imagine we have a model like this where it’s very sparsely activated, so different pieces of the model have different kinds of expertise And they’re called upon when it makes sense, but they’re mostly idle, so it’s relatively computationally [INAUDIBLE] power efficient But it can do many things And now, each component here is some piece of machine learning model with different kinds of state, parameters in the model, and different operations And a new task comes along Now you can imagine something like neural architecture search becoming– squint at it just right and now turn it into neural pathway search We’re going to look for components that are really good for this new task we care about, and maybe we’ll search and find that this path through the model actually gets us into a pretty good state for this new task Because maybe it goes through components that are trained on related tasks already And now maybe we want that model to be more accurate for the purple task, so we can add a bit more computational capacity, add a new component, start to use that component for this new task, continue training it, and now, that new component can also be used for solving other related tasks And each component itself might be running some sort of interesting architectural search inside it So I think something like that is the direction we should be exploring as a community It’s not what we’re doing today, but I think it could be a pretty interesting direction OK, and finally, I’d like to touch on thoughtful use of AI in society As we’ve seen more and more uses of machine learning in our products and around the world, it’s really, really important to be thinking carefully about how we want to apply these technologies Like any technology, these systems can be used for amazing things or things we might find a little detrimental in various ways And so we’ve come up with a set of principles by which we think about applying sort of machine learning and AI to our products And we’ve made these public about a year and a half ago as a way of sort of sharing our thought process with the rest of the world And I particularly like these I’ll point out many of these are sort of areas of research that are not fully understood yet, but we aim to apply the best in the state of the art methods, for example, for reducing bias in machine learning models, but also continue to do research and advance the state of the art in these areas And so this is just kind of a taste of different kinds of work we’re doing in this area– how do we do machine learning with more privacy,

using things like federated learning? How do we make models more interpretable so that a clinician can understand the predictions it’s making on diabetic retinopathy sort of examples? How do we make machine learning more fair? OK, and with that, I hope I’ve convinced you that deep neural nets and machine learning– you’re already here, so maybe you’re already convinced of this– but are helping make sort of significant advances in a lot of hard computer science problems, computer vision, speech recognition, language understanding General use of machine learning is going to push the world forward So thank you very much, and I appreciate you all being here [APPLAUSE] MEGAN KACHOLIA: Hey, everyone Good morning Just want to say, first of all, welcome Today, I want to talk a little bit about TensorFlow 2.0 and some of the new updates that we have that are going to make your experience with TensorFlow even better But before I dive into a lot of those details, I want to start off by thanking you, everyone here, everyone on the livestream, everyone who’s been contributing to TensorFlow, all of you who make up the community TensorFlow was open source to help accelerate the AI field for everyone You’ve used it in your experiments You’ve deployed in your businesses You’ve made some amazing different applications that we’re so excited to showcase and talk about, some that we get to see a bit here today, which is one of my favorite parts about conferences like this And you’ve done so much more And all of this has helped make TensorFlow what it is today It’s the most popular ML ecosystem in the world And honestly, that would not happen without the community being excited and embracing and using this and giving back So on behalf of the entire TensorFlow team, I really just first want to say thank you because it’s so amazing to see how TensorFlow is used That’s one of the greatest things I get to see about my job, is the applications and the way folks are using TensorFlow I want to take a step back and talk a little bit about some of the different user groups and how we see them making use of TensorFlow TensorFlow was being used across a wide range of experiments and applications So here, calling out researchers, data scientists and developers, and there’s other groups kind of in-between as well Researchers use it because it’s flexible It’s flexible enough to experiment with and push the state-of-the-art deep learning You heard this even just a few minutes ago, with folks from Twitter talking about how they’re able to use TensorFlow and expand on top of it in order to do some of the amazing things that they want to make use of on their own platform And at Google, we see examples of this when researchers are creating advanced models like Excel NAT and some of the other things that Jeff referenced in his talk earlier Taking a step forward, looking at data scientists, data scientists and enterprise engineers have said they rely on TensorFlow for performance and scale in training and production environments That’s one of the big things about TensorFlow that we’ve always emphasized and looked at from the beginning How can we make sure this can scale to large production use cases? For example, Quantify and BlackRock use TensorFlow to test and deploy BERT in real world NLP instances, such as text tokenization, as well as classification Hopping one step forward, looking a bit at application developers, application developers use TensorFlow because it’s easy to learn ML on the platforms that they care about Arduino wants to make ML simple on microcontrollers, so they rely on TensorFlow pre-trained models and TensorFlow Lite Micro for deployment Each of these groups is a critical part of the TensorFlow ecosystem And this is why we really wanted to make sure that TensorFlow 2.0 works for everyone We announced the alpha at our Dev Summit earlier this year And over the past few months, the team has been working very hard to incorporate early feedback Again, thank you to the community for giving us that early feedback, so we can make sure we’re developing something that works well for you And we’ve been working to resolve bugs and issues and things like that And just last month in September, we were excited to announce the final general release for TensorFlow 2.0 You might be familiar with TensorFlow’s architecture, which has always supported the ML lifecycle from training through deployment Again, one of the things we’ve emphasized since the beginning when TensorFlow was initially open sourced a few years ago But I want to emphasize how TensorFlow 2.0 makes this workflow even easier and more intuitive First, we invested in Keras, an easy-to-use package in TensorFlow, making it the default high level API Many developers love Keras because it’s easy to use and understand Again, you heard this already mentioned a little bit earlier, and hopefully, we’ll hear more about it throughout the next few days By tightly integrating Keras into 2.0, we can make Keras work even better with primitives like TF data We can do performance optimizations behind the scenes and run distributed training Again, we really wanted 2.0 to focus on usability

How can we make it easier for developers? How can we make it easier for users to get what they need out of TensorFlow? For instance, Lose It, a customized weight loss app, said they use tf.keras for designing their network By leveraging [INAUDIBLE] strategy distribution in 2.0, they were able to utilize the full power of their GPUs It’s feedback like this that we love to hear, and again, it’s very important for us to know how the community is making use of things, how the community is using 2.0, the things they want to see, so that we can make sure we’re developing the right framework and also make sure you can contribute back When you need a bit more control to create advanced algorithms, 2.0 comes fully loaded with eager execution, making it familiar for Python developers This is especially useful when you’re stepping through, doing debugging, making sure you can really understand step by step what’s happening This also means there’s less coding required when training your model, all without having to use session.run Again, usability is a focus To demonstrate the power of training models with 2.0, I’ll show you how you can train a state-of-the-art NLP model in 10 lines of code, using the Transformers NLP library by Hugging Face– again, a community contribution This popular package hosts some of the most advanced NLP models available today, like BERT, GPT, Transformer-XL, XLNet, and now supports TensorFlow 2.0 So let’s take a look Here, kind of just looking through the code, you can see how you can use 2.0 to train Hugging Face’s DistilBERT model for text classification You can see just simply load the tokenizer, model, and the data set Then prepare the data set and use tf.keras compile and fit APIs And with a few lines of code, I can now train my model And with just a few more lines, we can use the train model for tasks such as text classification using eager execution Again, it’s examples like this where we can see how the community takes something and is able to do something very exciting and amazing by making use of the platform and the ecosystem that TensorFlow is providing But building and training a model is only one part of TensorFlow 2.0 You need the performance to match That’s why we worked hard to continue to improve performance with TensorFlow 2.0 It delivers up to 3x faster training performance using mixed precision on NVIDIA Volta and Turing GPUs in a few lines of code with models like ResNet-50 and BERT As we continue to double down on 2.0 in the future, performance will remain a focus with more models and with hardware accelerators For example, in 2.1, so the next upcoming TensorFlow release, you can expect TPU and TPU pod support, along with mixed precision for GPUs So performance is something that we’re keeping a focus on as well, while also making sure usability really stands to the forefront But there’s a lot more to the ecosystem So beyond model building and performance, there are many other pieces that help round out the TensorFlow ecosystem Add-ons and extensions are a very important piece here, which is why we wanted to make sure that they’re also compatible with TensorFlow 2.0 So you can use popular libraries, like some other ones called out here, whether it’s TensorFlow Probability, TF Agents, or TF Text We’ve also introduced a host of new libraries to help researchers and ML practitioners in more useful ways So for example, neural structure learning helps to train neural networks with structured signals And the new Fairness Indicators add-on enables regular computation and visualization of fairness metrics And these are just the types of things that you can see kind of as part of the TensorFlow ecosystem, these add-ons that, again, can help you make sure you’re able to do the things you need to do not with your models, but kind of beyond just that Another valuable aspect of the TensorFlow ecosystem is being able to analyze your ML experiments in detail So this is showing TensorBoard TensorBoard is TensorFlow’s visualization toolkit, which is what helps you accomplish this It’s a popular tool among researchers and ML practitioners for tracking metrics, visualizing model graphs and parameters, and much more It’s very interesting that we’ve seen users enjoy TensorBoard so much, they’ll even take screenshots of their experiments and then use those screenshots to be able to share with others what they’re doing with TensorFlow This type of sharing and collaboration in the ML community is something we really want to encourage with TensorFlow Again, there’s so much that can happen by enabling the community to do good things That’s why I’m excited to share the preview of TensorBoard.dev, a new, free, managed TensorBoard experience that lets you upload and share your ML experiment results with anyone You’ll now be able to host and track your ML experiments and share them publicly

No setup required Simply upload your logs, and then share the URL, so that others can see the experiments and see the things that you are doing with TensorFlow As a preview, we’re starting off with the [INAUDIBLE] dashboard, but over time, we’ll be adding a lot more functionality to make the sharing experience even better But if you’re not looking to build models from scratch and want to reduce some computational cost, TensorFlow was always made pre-trained models available through TensorFlow Hub And today, we’re excited to share an improved experience of TensorFlow Hub that’s much more intuitive, where you can find a comprehensive repository of pre-trained models in the TensorFlow ecosystem This means you can find models like BERT and others related to image, text, video, and more that are ready to use with TensorFlow Lite and TensorFlow.js Again, we wanted to make sure the experience here was vastly improved to make it easier for you to find what you need in order to more quickly get to the task at hand And since TensorFlow is driven by all of you, TensorFlow Hub is hosting more pre-trained models from the community You’ll be able to find curated models by DeepMind, Google, Microsoft’s AI for Earth, and NVIDIA ready to use today with many more to come We want to make sure that TensorFlow Hub is a great place to find some of these excellent pre-trained models And again, there’s so much the community is doing We want to be able to showcase those models as well TensorFlow 2.0 also highlights TensorFlow’s core strengths and areas of focus, which is being able to go from model building, experimentation, through to production, no matter what platform you work on You can deploy end-to-end ML pipelines with TensorFlow Extended or TFX You can use your models on mobile and embedded devices with TensorFlow Lite for on device inference, and you can train and run models in the browser or Node.js with TensorFlow.js You’ll learn more about what’s new in TensorFlow in production during the keynote sessions tomorrow You can learn more about these updates by going to tensorflow.org where you’ll also find the latest documentation, examples, and tutorials for 2.0 Again, we want to make sure it’s easy for the community to see what’s happening, what’s new, and enable you to just do what you need to do with TensorFlow We’ve been thrilled to see the positive response to 2.0, and we hope you continue to share your feedback Thank you, and I hope you enjoy the rest of TF World [APPLAUSE] FREDERICK REISS: Hello, everyone I’m Fred Reiss I work for IBM I’ve been working for IBM since 2006 And I’ve been contributing to TensorFlow Core since 2017 But my primary job at IBM is to serve as tech lead for CODAIT That’s the Center for Open Source Data and AI Technologies We are an open source lab located in downtown San Francisco, and we work on open source technologies that are foundational to AI And we have on staff 44 full-time developers who work only on open source software And that’s a lot of developers, a lot of open source developers Or is it? Well, if you look across IBM at all of the IBM-ers who are active contributors to open source, in that they have committed code to GitHub in the last 30 days, you’ll find that there are almost 1,200 IBM-ers in that category So our 44 developers are actually a very small slice of a very large pie Oh, and those numbers, they don’t include Red Hat When we closed that acquisition earlier this year, we more than doubled our number of active contributors to open source So you can see that IBM is really big in open source And more and more, the bulk of our contributions in the open are going towards the foundations of AI And when I say AI, I mean AI in production I mean AI at scale AI at scale is not an algorithm It’s not a tool It’s a process It’s a process that starts with data, and then that data turns into features And those features train models, and those models get deployed in applications, and those applications produce more data And the whole thing starts all over again And at the core of this process is an ecosystem of open source software And at the core of this ecosystem is TensorFlow, which is why I’m here, on behalf of IBM open source, to welcome you to TensorFlow World Now throughout this conference, you’re going to see talks that speak to all of the different stages of this AI lifecycle But I think you’re going to see a special emphasis on this part–

moving models into production And one of the most important aspects of moving models into production is that when your model gets deployed in a real-world application, it’s going to start having effects on the real world And it becomes important to ensure that those effects are positive and that they’re fair to your clients, to your users Now, at IBM, here’s a hypothetical example that our researchers put together about a little over a year ago They took some real medical records data, and they produced a model that predicts which patients are more likely to get sick and therefore should get additional screening And they showed that if you naively trained this model, you end up with a model that has significant racial bias, but that by deploying state-of-the-art techniques to adjust the data set and the process of making the model, they could substantially reduce this bias to produce a model that is much more fair You can see a Jupyter Notebook with the entire scenario from end to end, including code and equations and results, at the URL down here Again, I need to emphasize this was a hypothetical example We built a flawed model deliberately, so we could show how to make it better But no patients were harmed in this exercise However, last Friday, I sat down with my morning coffee, and I opened up the “Wall Street Journal.” And I saw this article at the bottom of page three, describing a scenario eerily similar to our hypothetical When your hypothetical starts showing up as newspapers headlines, that’s kind of scary And I think it is incumbent upon us as an industry to move forward the process, the technology of trust in AI, trust and transparency in AI, which is why IBM and IBM Research have released our toolkits of state-of-the-art algorithms in this space as open source under AI Fairness 360, AI Explainability 360, and Adversarial Robustness 360 It is also why IBM is working with other members of the Linux Foundation AI, a trusted AI committee, to move forward open standards in this area so that we can all move more quickly to trusted AI Now if you’d like to hear more on this topic, my colleague, Animesh Singh, will be giving a talk this afternoon at 1:40 on trusted AI for the full 40 minute session Also I’d like to give a quick shout out to my other co-workers from CODAIT who have come down here to show you cool open source demos at the IBM booth That’s booth 201 Also check out our websites, developer.ibm.com and codait.org On behalf of IBM, I’d like to welcome you all to TensorFlow World Enjoy the conference Thank you [APPLAUSE] THEODORE SUMME: Hi, I’m Ted Summe from Twitter Before I get started with my conversation today, I want to do a quick plug for Twitter What’s great about events like this is you get to hear people like Jeff Dean talk And you also get to hear from colleagues and people in the industry that are facing similar challenges as you and have conversations around developments in data science and machine learning But what’s great is that’s actually available every day on Twitter Twitter’s phenomenal for conversation on data science and machine learning People like Jeff Dean and other thought leaders are constantly sharing their thoughts and their developments And you can follow that conversation and engage in it And not only that, but you can bring that conversation back to your workplace and come off looking like a hero– just something to consider So without that shameless plug, my name’s Ted Summe I lead product for Cortex Cortex is Twitter’s central machine learning organization If you have any questions for me or the team, feel free to connect with me on Twitter, and we can follow up later So before we get into how we’re accelerating ML at Twitter, let’s talk a little bit about how we’re even using ML at Twitter Twitter is largely organized against three customer needs, the first of which is our health initiative That might be a little bit confusing to you You might think of it as user safety But we think about it as improving the health of conversations on Twitter And machine learning is already at use here We use it to detect spam We can algorithmically and at scale detect spam and protect our users from it Similarly, in the abuse space, we can proactively flag content as potentially abuse, toss it up for human reviews, and act on it before our users even get impacted by it A third space where we’re using machine learning here is something called NSFW, Not Safe For Work I think you’re all familiar with the acronym So how can we, at scale, identify this content and handle it accordingly? Another use of machine learning in this space There’s more that we want to do here, and there’s more that we’re already doing Similarly, the consumer organization– this is largely what you think of, the big blue app of Twitter And here, the customer job that we’re serving is helping connect our customers with the conversations

on Twitter that they’re interested in And one of the primary veins in which we do this is our timeline Our timeline today is ranked So if you’re not familiar, users follow accounts Content and tweets associated with those accounts get funneled into a central feed And we rank that based on your past engagement and interest to make sure we bring forth the most relevant conversations for you Now, there’s lots of conversations on Twitter, and you’re not following everyone And so there’s also a job that we have to serve about bringing forth all the conversations that you’re not proactively following, but are still relevant to you This has surfaced in our Recommendations product, which uses machine learning to scan the corpus of content on Twitter, and identify what conversations would be most interesting to you, and push it to you in a notification The inverse of that is when you know what the topics you want to explore are, but you’re looking for the conversations around that That’s where we use Twitter Search This is another surface area in the big blue app that we’re using machine learning The third job to be done for our customers is helping connect brands with their customers You might think of this as the ads product And this is actually the OG of machine learning at Twitter, the first team that implemented it And here, we use it for what you might expect, ads ranking That’s kind of like the timeline ranking, but instead of tweets, it’s ads and identifying the most relevant ads for our users And as signals to go into that, we also do user targeting to understand your past engagement ads, understand which ads are in your interest space And the third– oh Yeah, we’re still good And the third is brand safety You might not think about this when you think about machine learning and advertising But if you’re a company like United and you want to advertise on Twitter, you want to make sure that your ad never shows up next to a tweet about a plane crash So how do we, at scale, protect our brands from those off-brand conversations? We use machine learning for this as well So as you can tell, machine learning is a big part of all of these organizations today And where we have shared interests and shared investment, we want to make sure we have a shared organization that serves that And that’s the need for Cortex Cortex is Twitter’s central machine learning team, and our purpose is really quite simple– to enable Twitter with ethical and advanced AI And to serve that purpose, we’ve organized in three ways The first is our applied research group This group applies the most advanced ML techniques from industry and research to our most important surface areas, whether they be new initiatives or existing places This team you can kind of think of as like an internal task force or consultancy that we can redeploy against the company’s top initiatives The second is signals When using machine learning, having shared data assets that are broadly useful can provide us more leverage Examples of this would be our language understanding team that looks at tweets and identifies named entities inside them Those can then be offered up as features for other teams to consume in their own applications of machine learning Similarly, our media understanding team looks at images and can create a fingerprint of any image And therefore, we can identify every use of that image across the platform These are examples of shared signals that we’re producing that can be used for machine learning at scale inside the company And the third organization is our platform team And this is really the origins of Cortex Here, we provide tools and infrastructure to accelerate ML development at Twitter, increase the velocity of our ML practitioners And this is really the focus of the conversation today When we set out to build this ML platform, we decided we wanted a shared ML platform across all of Twitter And why is that important that it be shared across all of Twitter? Well, we want transferability We want the great work being done in the ads team to be, where possible, transferable to benefit the health initiative where that’s relevant And similarly, if we have great talent in the consumer team that’s interested in moving to the ads team, if they’re on the same platform, they can transfer without friction and be able to ramp up quickly So we set out with this goal of having a shared ML platform across all of Twitter And when we did that, we looked at a couple product requirements First, it needs to be scalable It needs to be able to operate at Twitter scale The second, it needs to be adaptable This space is developing quickly so we need a platform that can evolve with data science and machine learning developments Third is the talent pool We want to make sure that we have a development environment at Twitter that appeals to the ML researchers and engineers that we’re hiring and developing Fourth is the ecosystem We want to be able to lean on the partners that are developing industry leading tools so that we can focus on technologies that are Twitter specific Fourth is documentation You ought to understand that We want to be able to quickly unblock our practitioners as they hit issues, which is inevitable in any platform And finally, usability We want to remove friction and frustration

from the lives of our team, so that they can focus on delivering value for our end customers So considering these product requirements, let’s see how TensorFlow is done against them First is scalability We validated this by putting TensorFlow by way of our implementation we called Deep Bird against timeline ranking So every tweet that’s ranked in the timeline today runs through TensorFlow So we can consider that test validated Second is adaptability The novel architectures that TensorFlow can support, as well as the custom lost functions, allows us to react to the latest research and employ that inside the company An example that we published on this publicly is our use of a SplitNet architecture and ads ranking So TensorFlow has been very adaptable for us Third is the talent pool, and we think about the talent pool in kind of two types There’s the ML engineer and the ML researcher And as a proxy of these audiences, we looked at the GitHub data on these And clearly, TensorFlow is widely adopted amongst ML engineers And similarly, the archive community shows strong evidence of wide adoption in the academic community On top of this proxy data, we also have anecdotal evidence of the speed of ramp-up for ML researchers and ML engineers inside the company The fourth is the ecosystem Whether it’s TensorBoard, TF Data Validation, TF Model Analysis, TF Metastore, TF Hub, TFX Pipelines, there’s a slew of these products out there, and they’re phenomenal They allow us to focus on developing tools and infrastructure that is specific to Twitter’s needs and lead on the great work of others So we’re really grateful for this, and TensorFlow does great here Fifth being documentation Now, this is what you would go to when you go to TensorFlow, and you see that phenomenal documentation, as well as great education resources But what you might not appreciate and what we’ve come to really appreciate is the value of the user generated content What Stack Overflow and other platforms can provide in terms of user generated content is almost as valuable as anything TensorFlow itself can create And so TensorFlow, given its widespread adoption, its great TensorFlow website, has provided phenomenal documentation for ML practitioners Finally, usability And this is why we’re really excited about TensorFlow 2.0 The orientation around the carrier’s API makes it more user friendly It also still continues to allow for flexibility for more advanced users The eager execution enables more rapid and intuitive debugging, and it closes the gap between ML engineers and modelers So clearly from this checklist, we’re pretty happy with our engagement with TensorFlow And we’re excited about continuing to develop the platform with them and push the limits on what it can do, with gratitude to the community for their participation and involvement in the product and appreciate their conversation on Twitter, as we advance it So if you have any questions for me, as I said before, you can connect with me, but I’m not alone here today A bunch of my colleagues are here as well So if you see them roaming the halls, feel free to engage with them Or as I shared before, you can continue the conversation on Twitter Here are their handles Thank you for your time Cheers [APPLAUSE] CRAIG WILEY: I just want to begin by saying I’ve been dabbling in Cloud AI and Cloud Machine Learning for a while And during that time, it never occurred to me that we’d be able to come out with something like we did today because this is only possible because Google Cloud and TensorFlow can collaborate unbelievably closely together within Google So to begin, let’s talk a little bit about TensorFlow– 46 million downloads TensorFlow has been massive growth the last few years It’s expanded from the forefront of research, which we’ve seen earlier this morning, to businesses taking it on as a dependency for their business to operate on a day in, day out basis It’s a super exciting piece As someone who spends most all of their time thinking about how we can bring AI and machine learning into businesses, seeing TensorFlow’s commitment and focus on deploying actual ML in production is super exciting to me With this growth, though, comes growing pains And part of that is things like support, right? When my model doesn’t do what I expected it to or my training job fails, what options do I have? And how well does your boss respond when you say, hey, yes, I don’t know why my model’s not training, but not to worry, I’ve put a question on Slack And hopefully, someone will get back to me We understand that businesses who are taking a bet on TensorFlow as a critical piece

of their hardware architecture or their stack need more than this Second, it can be a challenge to unlock the scale and performance of cloud For those of you who, like me, have gone through this journey over the last couple of years, for me, it started on my laptop Right? And then eventually, I outgrew my laptop, and so I had a gaming rig under my desk, right? With the GPU and eventually, there were eight gaming rigs under my desk And when you opened the door to my office, the whole floor knew because it sounded like [INAUDIBLE] Right? And but now with today’s cloud, that doesn’t have to be the case You can go from that single instance all the way up to a massive scale seamlessly So with that, today, we bring you TensorFlow Enterprise TensorFlow Enterprise is designed to do three things– one, give you Enterprise grade support; two, cloud scale performance; and three, managed services when and where you want them, at the abstraction level you want them Enterprise grade support, what does that mean? Fundamentally what that means is that as these businesses take a bet on TensorFlow, many of these businesses have IT policies or requirements that the software have a certain longevity before they’re willing to commit to it in production And so today, for certain versions of TensorFlow, when used on Google Cloud, we will extend that one year of support a full three years That means that if you’re building models on 1.15 today, you can know that for the next three years, you’ll get bug fixes and security patches when and where you need them Simple and scalable Scaling from an idea on a single node to production at massive scale can be daunting, right? Saying to my boss, hey, I took a sample of the data was something that previously seemed totally reasonable, but now we’re asked to train on the entire corpus of data And that can take days, weeks We can help with all of that by deploying TensorFlow on Google Cloud, a network that’s been running TensorFlow successfully for years and has been highly optimized for this purpose So scalable across our world class architecture, the products are compatibility tested with the cloud, their performance optimized for the cloud and for Google’s world class infrastructure What does this mean? So if any of you have ever had the opportunity to use BigQuery, BigQuery is Google Cloud’s kind of massively parallel cloud hosted data warehouse And by the way, if you haven’t tried using BigQuery, I highly recommend going out and trying it It returns results faster than can be imagined That speed in BigQuery, we wanted to make sure we were taking full advantage of that And so recent changes and recent pieces included in TensorFlow Enterprise have increased the speed of the connection between the data warehouse and TensorFlow by three times Right? Now, all of sudden, those jobs that were taking days take hours Unity gaming, wonderful customer and partner with us You can see the quote here Unity leverages these aspects of TensorFlow Enterprise in their business Their monetization products reach more than three billion devices– three billion devices worldwide Game developers rely on a mix of scale and products to drive installs and revenue and player engagement And Unity needs to be able to quickly test, build, scale, deploy models all at massive scale This allows them to serve up the best results for their developers and their advertisers Managed services As I said, TensorFlow Enterprise will be available on Google Cloud and will be available as part of Google Cloud’s AI platform It will also be available in VMs if you’d prefer that, or in containers if you want to run them on Google Cloud Kubernetes Engine, or using Kubeflow on Kubernetes Engine In summary, TensorFlow Enterprise

offers Enterprise grade support– that continuation, that full three years of support that IT departments are accustomed to– cloud scale performance so that you can run at massive scale, and works seamlessly with our managed services And all of this is free and fully included for all Google Cloud users Google Cloud becomes the best place to run TensorFlow But there’s one last piece, which is for companies for whom AI is their business– not companies for whom AI might help with this part of their business or that or might help optimize this campaign or this backend system, but for companies where AI is their business, right? Where they’re training hundreds of thousands of hours of training a year, petabytes of data, right? Using cutting edge models to meet their unique requirements, we are introducing TensorFlow Enterprise with white-glove support This is really for cutting edge AI, right? Engineering to engineering assistance when needed Close collaboration across Google allows us to fix bugs faster if needed One of the great opportunities of working in cloud, if you ask my kids, they’ll tell you that the reason I work in cloud AI and in kind of machine learning is in an effort to keep them ever from learning to drive They’re eight and 10 years old, so I need people to kind of hurry along this route, if you will But one of the customers and partners we have is Cruise Automotive And you can see here, they’re a shining example of the work we’re doing On their quest towards self-driving cars, they’ve also experienced hiccups and challenges and scaling problems And we’ve been a critical partner for them in helping ensure that they can achieve the results they need to, to solve this kind of generational defining problem of autonomous vehicles You can see not only did we improve the accuracy of their models, but also reduce training times from four days down to one day This allows them to iterate at speeds previously unthinkable So none of this, as I said, would have been possible without the close collaboration between Google Cloud and TensorFlow I look back on Megan’s recent announcement of TensorBoard.dev We will be looking at bringing that type of functionality into an enterprise environment as well in the coming months But we’re really, really excited to get TensorFlow Enterprise into your hands today To learn more and get started, you can go to the link, as well as sessions later today And if you are on the cutting edge of AI, we are accepting applications for the white-glove service as well We’re excited to bring this offering to teams We’re excited to bring this offering to businesses that want to move into a place where machine learning is increasingly a part of how they create value Thank you very much for your time today KEMAL EL MOUJAHID: Hi, my name is Kemal I’m the product director for TensorFlow So earlier, you heard from Jeff and Megan about the prod direction Now, what I’d like to talk about is the most important part of what we’re building, and that’s the community That’s you Sorry Where’s the slide? Thank you So as you’ve seen in the video, we’ve got a great roadshow, 11 events spanning five continents to connect the community with the TensorFlow team I, personally, was very lucky this summer, because I got to travel to Morocco and Ghana and Shanghai, amongst other places, just to meet the community, and to listen to your feedback And we heard a lot of great things So as we’re thinking about, how can we best help the community? It really came down to three things First, we would like to help you to connect with the larger community, and to share the latest and greatest of what you’ve been building Then, we also would like you– we want to help you learn, learn about ML, learn about TensorFlow And then, we want to help you contribute and give back to the community So let’s start with Connect So why connect? Well, first the community– the TensorFlow community has really grown a lot It’s huge– 46 million downloads, 2,100 committers,

and– again, I know that we’ve been saying that all along, but I really want to say a huge thank you on behalf of the TensorFlow team for making the community what it is today Another aspect of the community that we’re very proud of is that it’s truly global This is a revised map of our GitHub stars And, as you can see, we’re covering all time zones and we keep growing So the community is huge It’s truly global And we really want to think about, how can we bring the community closer together? And this is really what initiated the idea of TensorFlow World We wanted to create an event for you We wanted an event where you could come up and connect with the rest of the community, and share what you’ve been working on And this has actually started organically Seven months ago, the TensorFlow User Groups started, and I think now we have close to 50 The largest one is in Korea It has 46,000 members We have 50 in China So if you’re in the audience or in the livestream, and you’re looking to this map, and you’re thinking, wait, I don’t see a dot where I live– and you have a TensorFlow member that you’re connecting with, and you want to start a TensorFlow User Group– well, we’d like to help you So please go to tensorflow.org/community, and we’ll help you get it started So that next year, when we look at this map, we have dots all over the place So what about businesses? We’ve talked about developers What about businesses? One thing we heard from businesses is they have this business problem They think ML can help them, but they’re not sure how And that’s a huge missed opportunity when we look at the staggering $13 trillion that AI will bring to the global economy over the next decade So you have those businesses on one side, and then you have partners on the other side, who know about ML, they know how to use TensorFlow, so how do we connect those two? Well, this was the inspiration for launching our Trusted Partner Pilot Program, which helps you, as a business, connect to a partner who will help you solve your ML problem So if you go on tensorflow.org, you’ll find more about our Trusted Partner program Just a couple of examples of cool things that they’ve been working on One partner helped a car insurance company shorten the insurance claim processing time using image processing techniques Another partner helped the global med tech company by automating the shipping labeling process using object recognition techniques And you’ll hear more from these partners later today I encourage you to go check out their talks Another aspect is that if you’re a partner, and you’re interested in getting into this program, we also would like to hear from you So let’s talk about Learn We’ve invested a lot in producing quality material to help you learn about ML and about TensorFlow One thing that we did over the summer, which was very exciting is for the first time, we’re a part of the Google Summer of Code We had a lot of interest We were able to select 20 very talented students, and they got to work the whole summer with amazing mentors on the TensorFlow engineering team And they worked on very inspiring projects going from 2.0 to Swift to JS to TF-Agents So we were so excited with the success of this program that we decided to participate, for the first time, in a Google Code-in program So this is the same program, but for pre-university students from 13 to 17 It’s a global online contest And it introduces teenagers to the world of contributing to open source development So as I mentioned, we’ve invested a lot this year on ML education material, but one thing we heard is that there’s a lot of different things And what you want is to be guided through pathways of learning So we’ve worked hard on that, and we’ve decided to announce the new Learn ML page tensorflow.org And what this is a learning path curated for you by the TensorFlow team, and organized by level So you have from beginners to advanced You can explore books, courses, and videos to help you improve your knowledge of machine learning, and use that knowledge and use TensorFlow, to solve your real-world problem And for more exciting news that will be available on the website, I’d like to play a brief video by a friend Andrew Ng [VIDEO PLAYBACK] – Hi, everyone I’m in New York right now, and wish I could be there to enjoy the conference

But I want to share with you some exciting updates Deeplearning.ai started a partnership with the TensorFlow team with a goal of making world-class education available for developers on the Coursera platform Since releasing the Deep Learning Specialization, I’ve seen so many of you, hundreds of thousands, learn the fundamental skills of deep learning I’m delighted we’ve been able to complement that with the TensorFlow in Practice Specialization to help developers learn how to build ML applications for computer vision, NLP, sequence models, and more Today, I want to share with you an exciting new project that the deeplearning.ai and TensorFlow teams have been working on together Being able to use your models in a real-world scenario is when machine learning gets particularly exciting So we’re producing a new four-course specialization called TensorFlow Data and Deployment that will let you take your ML skills to the real world, deploying models to the web, mobile devices, and more It will be available on Coursera in early December I’m excited to see what you do with these new resources Keep learning [END PLAYBACK] KEMAL EL MOUJAHID: All right This is really cool Since we started working on these programs, it’s been pretty amazing to see hundreds of thousands of people take those courses And the goal of these educational resources is to let everyone participate in the ML revolution, regardless of what your experience with machine learning is And now, Contribute So a great way to get involved is to connect with your GDE We now have 126 machine learning GDEs globally We love our GDEs They’re amazing They do amazing things for the community This year alone, they gave over 400 tech talks, 250 workshops They wrote 221 articles reaching tens of thousands of developers And one thing that was new this year is that they helped with doc sprints So docs are really important They’re critical, right? You really need good quality docs to work on machine learning, and often the documentation is not available in people’s native languages And so this is why when we partnered with our GDEs, we launched the doc sprints Over 9,000 API docs were updated by members of the TensorFlow community in over 15 countries We heard amazing stories of power outage, and power running out, and people coming back later to finish a doc sprint, and actually writing docs on their phones So if you’ve been helping with docs, thank you, if you’re in the room, if you’re over livestream, thank you so much If you’re interested in helping translate documentation in your native language, please reach out, and we’ll help you organize a doc sprint Another thing that the GDEs help with is experimenting with the latest features So I want to call out Sam Witteveen, an ML GDE from Singapore, who’s already experimenting with 2.x TPUs, and you can hear him talk later today to hear about his experience So if you want to get involved, please reach out to your GDE and start working on TensorFlow Another really great way to help is to join a SIG A SIG is a Special Interest Group, and it helps you work on the things that you’re the most excited about on TensorFlow We have, now, 11 SIGs available Addons, IO, and Networking, in particular, really supported the transition to 2.0 by embracing the parts of contrib and putting them into 2.0 And SIG Build ensures that TF runs well everywhere on any OS, any architecture, and plays well with the Python library And we have many other really exciting SIGs, so I really encourage you to join one Another really great way to contribute is through competition And for those of you who were there at the Dev Summit back in March, we launched our 2.0 challenge on DevPost And the grand prize was an invitation to this event, TensorFlow World And so we would like to honor our 2.0 Challenge winners, and I think we are lucky to have two of them in the room– Victor and Kyle, if you’re here [APPLAUSE] So Victor worked on Handtrack.js, a library for prototyping hand gesture in the browser And then Kyle worked on a Python 3 package to simulate N-body,

to generate N-body simulations So one thing we heard, too, during our travels is, oh, that hackathon was great, but I totally missed it Can we have another one? Well, yes Let’s do another one So if you go on tfworld.devpost.com, we’re launching a new challenge You can apply your 2.0 skills and share the latest and greatest, and win cool prizes So we’re really excited to see what you’re going to build Another great community that we’re very excited to partner with is Kaggle So we’ve launched a contest on Kaggle to challenge you with question answering model based on Wikipedia articles You can put your natural language processing skills to the test and earn $50,000 in prizes It’s open for entry until January 22, so best of luck So we have a few action items for you, and they’re listed on this slide But remember, we created TensorFlow World for you, to help you connect and share what you’ve been working on So our main action item for you in the next two days is really to get to know the community better And with that, I’d like to thank you, and I hope you enjoy the rest of TF World Thank you [APPLAUSE]