Duration 28:31
16+
Play
Video

Better Understand Your Documents with AI by Elliott Ning, Architect, Google

Elliott Ning
Architect at Google
  • Video
  • Table of contents
  • Video
Video
Better Understand Your Documents with AI by Elliott Ning, Architect, Google
Available
In cart
Free
Free
Free
Free
Free
Free
Add to favorites
27
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Elliott Ning
Architect at Google

An experienced advisor of Google for enterprise cloud solutions, passionate about analyzing and solving key technological problems in machine learning, data analytics and infrastructure to bring business opportunity into reality. Has been working with top tier organizations by developing solutions and strategies on cloud and data center transformation. Holds Master's degrees in Software Engineering and Manufacturing Management.

View the profile

About the talk

Companies are sitting on a goldmine. You have all these documents - docs, emails, PDFs, forms, images… Some of these you need to understand as a requirement of doing business but others could give you valuable insight into your business and customers to help you make better decisions. But instead most documents are just sitting there, because it’s difficult to read, compare and understand the relationship between them. Document AI turns unstructured documents into structured data with the power to read, understand and make it useful. Under the hood, Vision and NLP technology give you the capability to unlock the values of documents and offers you a competitive advantage to re-invest & repurpose the resources for your organization. Key TAKEAWAYS1. Understand the possibilities of your documents2. Introduce the capability of Document AI technology3. Discuss Vision and NLP under the hood4. Explore how you can start

Share

Height one. Thank you for joining. My name is Aaliyah. I am responsible for Enterprise Cloud solutions by Google. Very happy to be here. I love sharing my mom age that you can learn and also discuss any cool design in my ears. I am a big fan and supporter of AI democratization. I hope we can bring more Innovative tools to the market. I'm one more people can choose the right technology to solve their problems with the power of the eye. So today let's talk about how we can solve a common problem.

in many organizations how to deal with growing number of document 1.3 knots is how we can understand them. But also turn them into meaningful information that can bring your values to your business. So here is my agenda today. Let's quickly talk about record Market challenges. And of course if there are charges there will be opportunities. Second I will briefly second level. Go over what is document AI? What is a do? And how can it help your business? And then we will uncover what is

under the hood. Capital technical talk about the technology including computer vision that's language processing and data warehouse. I'm going to back off. Summarize some of the key things that we cover. And also give some recommendations if we have enough time. All right. So let's first talk about the market. So we are in a rapidly evolving world. Customer expectations are higher than ever. As a business to stay competitive. You need to constantly look for ways to innovate. You get to bring new product new features

you service to the market. In order to be relevant and stay successful. So every company have documents. We're not just talking about papers, but also eating digital formats. So that's made watching order. Because according to the robotic process automation vendors the estimate Uber 50% off of Wicklow. beginning with the document there could be forms invoices Etc. So, how's that meeting us? We are facing challenges. The challenges are all these documents.

Do you have all kinds of documents? Warren's images contract nda's emails to the list of keeps going if you may have a manual process to understand because that will be the requirement. We're doing your business. Are there many others? So those documents could give you valuable insight into your business and your customers. What is that? Because it's difficult to read and understand. The complexity of document how to increase exponentially as business practice evolve. Hi Bond. The number of documents will not stop. You keep going.

The processes that we have involve large volumes of unstructured Contin may not be able to keep up with that kind of broke spacing High divers. So what's different new digital transformation tools at your organization about? the incoming data spends a wide variety of talking types and woman Have you talked to her at the Xcel? She flies PDF Json XML where a lot of people actually use freeform text. And there will be external knowledge. Play the modern business the market did Manning's new to evaluate arguments against not to show local environment.

muscle up Global Knowledge base It would be different kind of language knowledge localizations Behavior cultural and more. Do you want to costly the incoming data stream? Well multiple Hue General pneumatics. Someone change in a reference and impact many additional data records within your organization or maybe across multiple organizations. according to research only 20% off the Enterprise data. It's optimized for computers weed. So don't sorry about incomplete

easier to understand and utilize. the rest 80% of the data is not so they may not have a predefined formats wear patterns that you can easily Park information. They may require large amount of examples include. Where did we just you know some random conversation between your support your customers your people your engineers work there maybe images were logs that generated by some kind of systems that you built. So if you already have work yardage friends make some

of these documents. Of course, you can spend a lot of time and money on profiting and analyzing them. That's prone to electricity from Human errors or incomplete information. So what do we do? So what if there is a solution that you could easily processor document and a walker full battery? Without complex operations and large up from the message. That's what doc mini II comes to place. document Kearns unstructured documents into structured data It brings so much benefits for using AI to automate the process. Reduce

Tara my new process. You can think we've theoretically to any bottom as long as the infrastructure skills. if my sponsor it also boost productivity by having the capability to easily integrate with many Downstream services. And of course, you may understand better than human. So after all wheel of truck simulator If we can somehow magically turn all these documents into a structure format. That means we can Now understand and you are in the way that we prefer.

We can run queries on it. The dark mini I is not a single product. He's actually a collection of tools and services that can help you in different stage of the process here. It's a technical document understanding. That one is to weed and extract the text and values from the documents. Yeah, you may think about where dogs right? Yes. Text me back is relatively easier, but what about images? What about PDX? biotin skin including nytimes the key for stage one is to make sure that we have the right

information collected from the wall format. The two ways to understand information extracted from documents. Are there many ways including summarizing? So what if I keep counting up, all these documents are talking about? Classifying or they'd long into specific categories like Market should product worse be sure that you care. Analyzing you can analyze things like sentiment. Are they cutting a happy or not when they talking about certain things? And then step is to make it useful by creating meaningful inside. What is mean by meaningful insight?

Defensive data is a structure form. We can do all kinds of analytical operation like aggregating counting sweating Group pain center. Play trampled by you can find out how many customers are happy about something not happy about something may be able to cross-reference customers and different product Market segment to compare the behaviors. The document a guy gives you a competitive advantage. And allows you to reinvest repurpose your resources. You can increase your operational efficiency by 40 automating the process. Improve your customer experience

and one pointy unlock the value by providing strategic insights from the documents, which you are not leveraging today. I can see talk to me. I have been using the money industry use case. We tell Healthcare Financial entertainment Manufacturing. But you make a Wandering. Okay now. It's really useful. But how does it work? How can I start? Now looking Bill what is under the hood of document AI? Y'all are some core components. So it's not limited to only these technology.

Word document extraction computer vision is the key. It will be optical character recognition OCR. She'll recognize from the images. And of course, it will be different type of specialized posters or form and invoices. War information understanding we can leverage natural language processing are using the previous AGI on generic use case where you can build customized models. I taking the advantage of Ottawa now and I'll p with minimal effort and machine learning activities. Atlanta, of course you going to need a solid data warehouse to store and

getting inside from the analyze data? You can choose to build everything yourselves. What are in the Pac-12 consisting different kinds of men's tools and giving you the flexibility to create a pipeline that you want? My suggestion is always a used white LG to make more sense to your Christmas. Don't know to read the blind the documents properly you need computer vision or OCR. Turn the document into text women. Because computer can we text? How many people actually willing to

build OCR by yourself? At least not many because not very productive. So here's the API service that you can create with your processor with literally two clicks. I'm Christine created it will be important that you can send your files to and get results back very very quickly. So there are many title processor is that you can create based on your documents and applications. Are two main categories general-purpose and specialized Potters. Where General documents you can use things like basic

ZR2 identify and extract text in different type of document. Like the one that's showing on flight. It has Wilcox accuracy. It supports over 200 languages. And also speak a different language for handwriting recognition. For domain specific documents. You can choose from different title for Masters like the 209 10:40 Central to extract a special sculptures from the documents with Kevo care. There's also English Potter's which can extract text and dahlias from invoices State addresses phone numbers

companies price and other Salvage. Now you have to all the konshen in text format. Let's try to understand it. So in order to understand the text continent. We need Anarchy. The fastest way to achieve an LP is through NLP API. So there are powerful free build models of natural language that empowers developers to easily apply natural language understanding your applications with features, including Central analysis and send text without YouTube build any models.

So this is the design of our model your data. Do you send your prediction data and you can get results back right away? But if you say I need a high-quality in customized model. And I would suggest you take a look at all to our mouths. Automl natural language enables you to build and deploy custom machine learning models that analyze documents by categorizing identifying attitudes. So instead of spending large amount of time and investment on acquiring lot of images identifying training models.

You can just let Auto amount it handle it behind the scene. You wish to go through the whole machine learning training process. But in a much more efficient way. So what you need to do is to Peter tax accountant into the system. You can label keywords and phrases on the UI for 2 API. So awful amount helps you automatically determines the best model for the data. There's no need for tuning. Because also no will also test any parallel multiple combinations of models and barometers and make sure get the best result.

What's the trading price of complete atin point will be available for you to evaluate tomorrow manually? Once you confident about the custom, I'll let you build. You do one click to deploy and integrate with your existing applications. Without one word about anything like hosting High then ability scaling in a fully managed fashion. It sounds too. Good to be true AI subject. The biggest reason why I support a I can walk you to let more people accessible to useful.

So there are many benefits for using aqua Mount natural language or document ER. You get custom motor. You trained custom machine learning models with much faster time. It can be done in hours instead of days in weeks. The only requires less amount of data hundreds examples. Inside of a brightly I which requires Millions examples. Integrated my Sleepy Eye. the interview with accessible via comprehensive ATI you can send text via request storage. Is Power by Google's offline. Models? No leverage Google state-of-the-art technology to produce

high-quality models. So instead of building from ground up your building on top of a sophisticated based Foundation that many Google machine. Things you can do you can have custom content classification can create labels to customize models for unique. You can get custom entity extraction and identify entities within document and label them based on your future names. You can also do a custom sentiment analysis. Understand the overall can write feelings or attitudes expressed in words, but also in a block of text.

Make it to knit your own domain specific sentiment scores. Drake so in order to make protected information useful you going to need a solid state of addition? a good data warehouse Hey Corey, it's Google's fully-managed. How do you spell aboard in a container Warehouse solution? Be quiet and very easy to use. SQL queries It's herbalist. There's no hosting the system pressure to worry about. Performance is amazing. You can skilter bite infection and petabyte. It supports both batch and streaming. Sony offers a real-time Insight from your

streaming data They have built-in machine loading out of the box or things like me questions and recommendations. It works with high-speed England Marie Bing or fast recording and an ellipsis. Be Crazy Design to make data analyst and it is secure offering that data encryption Always by default to ensure data security. Rocori it's a very sophisticated and mature service with many moving pieces the high level. Storage and compute are decoupled. infant scale independently on the map this offers immerse

flexibility and also cost control for a business. Because you don't need to keep your expensive computer resource up and running all the time. So this is a very different than the traditional. No bathe data warehouse solution or any on-prem systems. On the storage side. The system is called Colossus. Which is a successor to Google boxes BFF. How do you remove? It's incredibly for point and super scalable in the distributor. There's lots of optimization that would be done to optimize the data distribution for your query performance.

On the computer side. The execution engine is called Trammell. The Dremel engine use a multi-level serving free for filling out your oil SQL queries. Italian clear signs law to queries as on demand basis. So it's single user to get thousands of computers lots to run Corey and regardless of the note l e words. So due to the separation design by what are to connect the compute and storage layers Bakery use ultra V Network which hand-deliver terabyte of data in second directly from Storage in to compute

for running the Dremel jobs. They're just so much to do and I can talk about where was amazing technology. So feel free to read more about the documentation where you can connect with me to discuss any design. All right, so I think we have briefly walk through a life of document. First you can convert input right like PDF to images. Because use optionally ultramel Vision something I actually didn't talk about because of time but it is available to classify images into domain specific types and categories.

And then you can run OCR with vision API from the documents. Of course if they are ready text face woman. After that, you can use a trained off Hormel natural language model to extract the key if you classify specific domain and analyze attitudes. Finally can see all the results in Vickery. Now. You can run SQL queries to analyze and also on this information to to the way that you want. Forgot many components. You can mix a mouse to build a prototype by Journey. Depending on the news case files and most important thing is the Insight that

you are expected to get from this. So that pretty much wraps up my presentation today my suggestion for everyone. If you do have a whole bunch of document think about what my duty is to get a high-level understanding what they are what you may be able to get out of it and then you can go to Google Cloud website search for document a I take a look at the available tools. You don't need to be machine learning experts and very minimum coding required. You can start thinking and even trying the activities like that. You would need a long the document understanding process. What

if you have any additional questions Google team is always willing to help. Thank you, babe.

Cackle comments for the website

Buy this talk

Access to the talk “Better Understand Your Documents with AI by Elliott Ning, Architect, Google”
Available
In cart
Free
Free
Free
Free
Free
Free

Ticket

Get access to all videos “Global Artificial Intelligence Virtual Conference”
Available
In cart
Free
Free
Free
Free
Free
Free
Ticket

Similar talks

Hudson Mahboubi
Sr. Manager Of Data Science at Workplace Safety & Insurance Board
Available
In cart
Free
Free
Free
Free
Free
Free
Sudip Chakrabarti
Venture Investor at Madrona
+ 4 speakers
Ian Foley
Venture Partner at Level Ventures
+ 4 speakers
Anupam Rastogi
Partner at Emergent Ventures
+ 4 speakers
Vik Sasi
Partner at Dreamers VC
+ 4 speakers
Dennis Liu
Investor at Ford
+ 4 speakers
Available
In cart
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Better Understand Your Documents with AI by Elliott Ning, Architect, Google”
Available
In cart
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
561 conferences
22100 speakers
8257 hours of content