Having spent the last couple years building AWS SageMaker, I love ungating the power of data and driving improved decision making through the application of machine learning and advanced analytics. Whether leading analytic teams tasked with driving business metrics or building tools/platforms so that others can do analytics that they previously thought impossible, I seek to create more optimal decision making for my team and the world.My passion is driving corporate profitability and fueling sustainable business growth for multi-billion dollar companies. I am a highly-analytical and growth-centric product visionary and business strategist with a consistent record of success. I am committed to exceeding top and bottom line growth through data-driven decision making that aid C-level executives.A few accomplishments include:==> I drove substantial incremental profit and margin improvement for Amazon.==> I significantly improved Free Cash Flow in year one in the role.==> I increased conversion and growth rate through improved customer interfaces and improved marketing targeting.Recognized as an expert in building or enhancing systems for analyzing and interpreting sophisticated data and metrics while steering revenue and profit growth for Amazon.com, I have received several vertical promotions for the last 7 years.View the profile
About the talk
Learn how customers have unlocked the power of data by utilizing Google’s AI Platform. From APIs to AutoML to writing your own model code, see real-world examples of how customers create value, and critical tips on how to accelerate your own AI journey.
Speaker: Craig Wiley
Google Cloud Next ’20: OnAir → https://goo.gle/next2020
Subscribe to the GCP Channel → https://goo.gle/GCP
product: Cloud AutoML, AI Platform; fullname: Craig Wiley;
event: Google Cloud Next 2020; re_ty: Publish;
Hi, my name is Craig on the director of product management for cloud, AI platform here, at 4 today, to be talking to about how to create value with the breadth and depth of our a class. Creating value with the AI platform. Simply comes down to putting the best of Google's ad Technologies, to work for you and for your company. What is this mean? Fundamentally this mean how can we knock your business strategy to the right way? I absorbed and strategy will talk about the different ways. The Google helps you deploy a without compromising on flexibility speed quality or scale.
And finally, we'll talk about some of my customers that have been transformed by tapping into the Innovation from Google. If I think about AI fundamentally, I only does two things. One it helps you grow. Your Market increased subscribership increase, users increased their spend or increase their conversion, or it helps you in the back end Drive efficiencies reduce costs drive out, waste, from the system. Together, these two capabilities allow you to achieve previously
unimaginable results for you and for your company using this critical technology. Is not limited to this business. So that industry opportunities to use a. I can be found in every vertical whether its retail our capabilities in healthcare Financial Services. Things like a lending IR, anti-money-laundering media and entertainment or industrial manufacturing with things like visual inspection capability or public sector. The rich with opportunities, all of these are areas that we can work on.
As you think about AI investments, in AI absorption. When you think about what Google offers, you can really break it down into three levels of investment. The first is a, i out of the box. This is very low investment very quick to Market, but doesn't come with a lot of customization II is deploying customer. On the other hand does allow for customization takes a bit more effort, but is still very quick to Market and last slave is building end-to-end AI. This is data scientist. You work with investors to build your own, bespoke
models for areas where this will be critical to your company's strategic Innovation or your differentiation in the market. We can think about these three levels with this example. The first example of kind of d a i right out of the box would be something like optical character recognition. You could imagine simply calling our vision and easily read the documents that you may have. Second would be something like potentially content classification. This is an area where the New York Times invested with us on using automl to quickly and easily
classify their entire Content Library. And then finally, we come to actual pure Insight generation deep Investments that provide unique benefit, this can be seen wonderfully with kaggle an organization that we run that has competitions for data scientists from the academic literature around. These three levels of understanding can be seen in our products as well. At the first layer, we see pre-trained API, these require no real work from you. All you need to do is integrate with the API, no data science is needed from you. Instead, you can simply get your image label.
Get your document understood the extraction of document, or in this case you can see across frames per second would be customary amount with this capability sadar models. Don't have the objects. You need labeled you want specific fish. Not just what species it is. You can take your own photos quickly and easily labeled up with our service and then move them in and train your own AI model with no code. And finally, for those more powerful model or more interesting, bespoke use cases, we offer the full end-to-end capability with our
core tools. You can think of it this way across the top of the stack, we have our applications vision and video conversation. Language and structured data. So whether it's labeling image, you're working with video, whether it's speech to text, text to speech or setting up. A conversational bot translation or entity extraction or sentiment analysis. You have all of your needs with our language capabilities. And finally was structured data, you can build models on your structured data without python her. Any sort of coding require then Benito's
applications, we have the cord tools designed to help their data. Scientists build, a powerful models. They want to explore and do whether it's notebooks as a service day labeling training as a service prediction as of service, all of the critical components, you'll need to ensure that you are building the best AI capabilities coming example. Later this year will have new improved work clothes for auto amount within the console. Experience no longer, requiring make specific decisions and choices. But, instead, making it all
easy, right out of the box. You can see that with our stack you can start by creating a dataset and then quickly and easily with no code required, move all the way through to a model deployment, putting your model into production ready to be called in a highly auto-scaling environment. Now, for those that wish to do something more sophisticated or wish to build their own model, they'll have the full capability to do. So, as well. If we move Beyond automl, one primary area investment for us, has been machine learning Ops or ml Ops in this
area. Nothing has been as exciting to us, as what we've seen customers. Do with our managed pipelines, continuous production and meditate to help make decisions customers will now be able to support both a fully CI CD capability as well as to start creating pipelines. That will support much more sophisticated operations such as quickly, and easily retraining your model once. These pipeline. What they do is, they allow you to take your monolith of of code that you use to build this model and break it down into each of the
individual components. Now, it's much more easily be stood as well as pipelines down into your model to use for things like extracting data validating data or evaluating the model, these capabilities significantly reduce the time of experimentation. We had one customer telling us that they have increased the rate of experimentation by 7x. Using this technology, we couldn't be more excited to see how your company will use this. Find me as is critical to any service of this type metadata management is, is a key piece of what we say we believe is needed to effectively run, AML option, buyer,
whether that's automatic meditative login from our hosted services or custom metrics that you may want to divide both are quickly and easily captured for things like compliance, MLS continuous training, and continuous delivery, or experimentation development model, an artifact comparisons, as well as improved ML and explain about the compliance story. It allows you to record All you have to do is remove the customer from the dataset in question or from all of the data fits that they're in. And then with pipeline than ml,
just by determining the fact that the dataset hadn't changed automatically retrain the model and redeploy, the model Network, We're proud of our AI platform platform. Do things really well. One users where they are too quickly. Deploy is the highest quality I and three allows you to scale and streamline not AI in unbelievable way. Meeting customers where they are. What does that mean? Well, as you can see here, I think we all fall somewhere on the spectrum of skills and capabilities within machine. Learning and whether you're someone who's just starting out
or whether you're someone who's already comfortable with the most sophisticated tools, we have solutions for you. When we talked about meeting customers, where they are two companies that are interesting to call out. The first is Cruz, Cruz is an autonomous vehicle company making tremendous strides and moving to a world where my son will never have to get his driver's license. Winston Cruze build models. And we have worked closely with them on optimizing their environments and optimizing their tensorflow capabilities to decrease their training times. So that they can more quickly and
more easily cycle through experiments and move for the world in which cars are fully autonomous. Now, on the other side is the city of Memphis. The city of Memphis not traditionally known as a tremendous machine learning power out. But in this case, they were actually building a model very similar to ones that Cruz has no doubt. Had Bill the city of Memphis wanted to understand where they had the most problematic potholes. So they used auto ml to build a Popple detector, and sure enough were able to save $20,000 a year in the reduction in claims cost.
As you can imagine these two customers come with very different skills and expertise and both were able to find tremendous value with Google and its platform. We do see, as I said to meet customers where they are. So for example, Rai notebooks not only do they come free prepared for the use of classic traditional ml framework such as tensorflow or what, they also have data analytics capabilities so that you can get a big data image with things like Apache being pre-installed. So you can quickly
Explain ability is at the heart of understanding our machine learning. We Believe heavily and have invested heavily in this space. As you can see, at the top of the light or images of explainability showing what pixels in particular, the model may have used to render its judgment Down Below. On the slide art is a picture of feature importance from a tabular use case both of these provide tremendous value to data scientist in both understand why the model is doing what it's doing as well as ensuring that the metal is doing what
it's supposed to be doing. We don't think of explainability unexplainable AI is some separate field that you know can only be used by the experts. We believe it's Central and as such we included with every prediction in our production service, we simply publish all of the explainability results in the same logs. We publish your inferences so that you can go back and look to see if you're experienced an unexpected dress for skew. So that you'll know when it's time to retrain the mobile. Not only do we like to meet customers where they are, but we ensure that we can quickly
deploy, the highest-quality, ai4 easy. This might be in the area of sight language conversation or with structured data. In the area of sight or excited to work with box boxes, using our vision API to help their customers manage and gain Insight from their image files and speed up image Centric processes and workflows. Whether it's the work that box is doing, or whether its image and videos search or industrial inspection. Our vision capabilities are unparalleled in the market. Lester bi
translation, the natural language capabilities weather at the entity extraction, document classification, sentiment analysis, or in this case translation Bloomberg is a company that sells information so that information in whatever language their customers is a critical capability. Imagine how breaking down the side of languages might be beneficial to your bit. Next with speech, voice and conversational boss, whether it's speech to text text to speech, video transcription, voice commands. All of this is
made possible with these capabilities, and we're excited about the work we've done with Domino's, using dialogflow, so that they could meet all of their goals in Milestone as they quickly and easily, deployed and model to allow customers to quickly and easily order through the use of a conversation. Find a thumbs-down to structured data structured data is what many of our business is wrong on. Whether it's in our data warehouse, in the business systems, we may have running Now usually when we talked about machine learning, we talked about very complex and
kind of very important use cases sometimes though. It's okay for a i to be fun and this is one of those models. We partnered with Fox Sports and tables. Capability are structured data capability to build a model that would predict when the next Wicket would follow. For those of you who are Cricket, fans are for those who are this. You know, may not seem all that exciting but I will say that they increased their usage. They also increased viewership as well as reduced the number of critical cost metrics and customer tainment because customers were so
interested in. Now all of a sudden when I'm watching, I can simply pay attention to the part of the match that I know is going to be relevant. Or I may simply want to tune in to see how well the I actually predicts the outcomes. Finally, the ability to scale and streamline with these, this fundamentally comes down to Google passing over, many of its symptoms, so that you can use them in your environment. So that you have the same power of scale, brings to the table when
joining gcp realized that they wanted to sort through an index /. 40000000000 images, working with us, we were able to help them increase efficiency by 15% in doing this. As I mentioned before, AI notebooks are critical. They that many of our users interact with our services. I can think back to the days when managing a machine-learning team and how many scientists at the end of the day when pack their laptops into their backpack and head home for the evening, with their notebooks saves locally. And probably a significant
amount of data saves locally with our Cloud notebooks available to you, that's never again a concern, our Cloud notebooks come with tremendous levels of security, including data exfiltration capabilities, you know, that both your coat and your data is safe when your employees are using Cloud Ai and our AI platform notebooks, Finally. AI platform prediction service platform prediction service allows you access to some of the most cutting-edge Hardware things like in
videos, T4 gpus, it automatically laws, all of your inferences. The incoming payload of information as well as those explanatory variables and we log all of that and big Prairie so that you have quick and easy access to doing some of the analysis on it. Finally, it's scales automatically with traffic. All you have to do is tell us what kind of processor you want, running it and how high up you would want us to scale. If the traffic comes in, as the traffic comes, we will automatically add those resources. So that you're only paying when you actually are seeing that traffic. As I
mentioned before a, i platform pipelines is a transformational technology for many teams. Being a 7x increase in the rate of experimentation. All of a sudden makes problems taken me months, may take me only a week with the power of that kind of velocity increase. You can only imagine the capabilities that you're a machine learning or data science. Team may be able to effect. Tensorflow, enterprise tensorflow, Enterprise is uniquely available on Google Cloud,
three things. First of all, of gives you cloud-scale performance with highly optimized fineries to allow things like increase iOS be so that you can see improved training or production also it's used in all of our men services. So you know that if your training as a service you're protected by tensorflow Enterprise, I say protected because the third aspect is support, many of us have many have concerns about using open-source in production. Can you trust that it will be well supported with security patches and Bug fixes?
We're excited to say the enterprise, we extend the one year of support. The tensorflow gives two three full years of support so that you your it department and your business can be comfortable insurance. This software is fully protected from security and Bug fixes for the full duration of the life of your mama. All of this comes back into the II platform, whether it's the applications for vision conversation language and structured data or whether it's the core Services of notebook is a service or training is in
service or pipeline that you may want to use all of it. We are excited to partner with you to deal with models that go beyond your expectations. Finally, to wrap it all up. First map, your business objectives to irrelevant. AI strategy how this technology can help you figure out what you need help with him. And let's figure out the right technology to use second partner with a company that has tremendous experience deploy in a i at scale, we have learned from our mistakes. We have learned them already.
You don't need to relearn finally plan for the end-to-end AI life cycle Beyond, just the model training by using pipelines. You have the capability to create a full CI, CD system that will relearn, as soon as the results. Google has been called out by a number of organizations for a leadership in machine learning. You can see her from in Foresters way there for their computer platforms that they've named. Google is a leader. Similarly, in Gartner's magic. Quadrant Cloud developer Services magic. Quadrant Google was named a leader in 2020. With the wards across
the industry. We are very proud of what we have built and very excited to work with customers. Like you on building and deploying new and exciting with that. I would like to allow one of our customers to speak for a moment on how they've unlocked the value of a i with Google Cloud. Cardinal Health has about 50,000 employees across 26 countries. Our mission is to be essential to health care. We provide products and services so that the healthcare providers can focus on their patients. The industry has evolved over the last five to ten years in the world is basically moving
to personalization machine learning and artificial intelligence. And I really just had a great connection culturally work ethic than also have some coming up vacation with Google cloud in Google Cloud everyday. We have 2500 emails coming in from the customer through our pass. AP CRM system with different messages, different intentions. So we have to send it to the human routers, a group of 40 people who are sorting and categorize the email before we send it to the agents, to perform the tasks,
that is a very labor-intensive process. We leveraged Google Cloud native services and create 8, mL bottle, do that automatically sorted out our emails, we use the human categorize emails. So we aren't all that categorisation can be found in machine learning and then with deployed armaldo into Google Cloud. The people that we afraid of through the machine learning, they move to do the actual work rather than sorting the emails. We are very happy about this project because it's a milestone for us. We proved that
a touchy verbal and its really improve customer service. Rai platform of choice is Google Cloud. We're very lucky to have to work with Cardinal Health and we're very lucky to see some of the solutions that they've been able to do with that you're some resources. If you're interested in, would like to learn more about our AI products or about our building use AI capabilities Thank you for your time today. And I look forward to meeting you and hopefully working with you on deploying exciting II for your
Buy this talk
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.