Como usar tech para facilitar dev e trazer beneficios para o cliente?
[Cover at high level + highlight our customer-focused approach – no need to address everything on this slide]
Our approach for ML is similar to how we approach other areas of the AWS business:
We focus on customers’ business needs and developer capabilities
Innovate rapidly on behalf of our customers
Offer a broad and deep set of services for our customers
We are also focused on providing customers with choices – which is why we support the most popular frameworks.
Our mission is to enable customers to transform their businesses by putting ML in the hand of every developer. [Note: adding a customer-focused angle to our mission will be most impactful with the analyst audience]
Roadmap baseado no cliente (90%+), Uma enorme variedade de serviços em produção
[Swami: Recommend calling out a few customer names, which AWS AI/ML service(s) they are using and how]
Below are the customers Andy is going to call out in his keynote:
LIBERTY MUTUAL
SLACK
C-SPAN
INTUIT
PINTEREST
CAP ONE
AMERICAN HEART ASSOCIATION
YELP
FINRA
NBC
-------------------------------------------------------------------------------
Parking lot
1/ FINRA is a not-for-profit organization authorized by Congress to protect America’s investors by making sure the broker-dealer industry operates fairly and honestly.
2/ FINRA receives millions of documents with unstructured data to support investigative, examination, and compliance processes. Previously, investigators and examiners had to manually go through documents page by page to find what they needed.
3/ FINRA uses Amazon Comprehend to quickly extract names of individuals and organizations, and match extracted entities to FINRA records, flag individual of interest, generate summaries, and detect similarities with other documents.
4/ This exponentially increases the scope of investigations while shortening the time to conclusion.
9/ Sky News is a 24-hour international multimedia news organization based in the UK.
10/ They wanted to identify celebrity guests at the Royal Wedding in real time, to improve the experience for viewers of their broadcast.
11/ They used the GrayMeta Platform powered by Amazon Rekognition to deliver the “Who’s Who Live” feature in the broadcast.
12/ This provided viewers with biographical information of identified guests as they arrived at the event.
13/ NBC is a broadcast television network with over 200 stations and affiliates.
14/ Content that is broadcast on TV must get screened for compliance per the rules of the network, time of day and region of the world. The same content must also be screened for proof of advertising content. The labor needs and overall cost to screen nonstop content across 50 channels is high, as are the fines for noncompliance.
15/ NBC uses Amazon Rekognition (Moderation, Labels, Text, Faces, Celebrity), to conduct an initial review of all content before it is broadcast and flag potential issues. Human screeners then review any flagged content to confirm the issue and take action.
16/ This focuses the efforts of NBC’s human screeners, lowering the total cost and shortening the review cycle.
17/ MasterCard NuData Security, A MasterCard Company, is trusted by some of the largest global brands to verify online users.
18/ Fraudsters are increasingly using sophisticated automation and other techniques to circumvent even multi-factor authentication.
19/ With its NuDetect behavioral biometrics solution, NuData tracks hundreds of behavioral attributes such as typing cadence, the angle a device is held, pressure, device settings, and how the user navigates through your website or mobile applications. With this data, they use Amazon SageMaker to create and deploy machine learning models based on hundreds of billions of data points that help flag fraud in real time.
20/ This means their approach to fraud detection is continually adapting, changing, and evolving to keep pace with fraudsters.
We see the Machine Learning stack having three key layers.
ML Frameworks:
The bottom layer is for expert machine learning practitioners—researchers and developers.
These are people who are comfortable building models, tuning models, training models, figuring out how to deploy into production, and manage them themselves.
And the vast majority of machine learning in the cloud today at this layer is being down through Amazon SageMaker which provides a managed experience for frameworks, or the AWS Deep Learning AMI that we built that effectively embeds all the major frameworks.
Infrastructure:
AWS offers a broad array of compute options for training and inference with powerful GPU-based instances, compute and memory optimized instances, and even FPGAs.
Our P3 instances provide up to 14 times better performance than previous-generation Amazon EC2 GPU compute instances.
C5 instances offer higher memory to vCPU ratio and deliver 25% improvement in price/performance compared to C4 instances, and are ideal for demanding inference applications.
We also have Amazon EC2 F1, a compute instance with field programmable gate arrays (FPGAs) that you can program to create custom hardware accelerations for your machine learning applications. F1 instances are easy to program and come with everything you need to develop, simulate, debug, and compile your hardware acceleration code. You can reuse your designs as many times, and across as many F1 instances as you like.
The new Amazon EC2 P3dn instance has four-times the networking bandwidth and twice the GPU memory of the largest P3 instance, P3dn is ideal for large scale distributed training. No one else has anything close.
P3dn.24xlarge instances offer 96vCPUs of Intel Skylake processors to reduce preprocessing time of data required for machine learning training.
The enhanced networking of the P3n instance allows GPUs to be used more efficiently in multi-node configurations so training jobs complete faster.
Finally, the extra GPU memory allows developers to easily handle more advanced machine learning models such as holding and processing multiple batches of 4k images for image classification and object detection systems
ML Services:
But, if you want to enable most enterprises and companies to be able to scale machine learning, we’ve solved that problem for organizations by making ML accessible for everyday developers and scientists. Amazon SageMaker removes the heavy lifting, complexity, and guesswork from each step of the machine learning process.
SageMaker makes model building and training easier by providing pre-built development notebooks, popular machine learning algorithms optimized for petabyte-scale datasets, and automatic model tuning, enabling developers to build, train, and deploy models in a single click.
SageMaker is already helping thousands of developers easily get started with building, training, and deploying models.
AI Services:
At the top layer are AI services which are ready-made for all developers—no ML skills.
For example, customers say here is an object, tell me what's in it, or here's a face, tell me if it's part of this facial group using Amazon Rekognition
Or let me translate text to speech using Amazon Polly
Or let’s build conversational apps with Amazon Lex.
Convert speech to text with Amazon Transcribe
Translate text between languages using Amazon Translate
Understand relationships and find insights from unstructured text using Amazon Comprehend
AI Services:
AI Services are intentionally easy to use. They can be accessed via a simple API call.
We’ve pulled the best and most targeted capabilities into ready-made services--for example image recognition or transcription.
The focus here is really on enabling any developer—no ML skills required—to be able to develop AI applications using one of our services.
These API services, used in conjunction, create compelling solutions that really target business problems and use cases.
Customers can build these capabilities into their new and existing applications to reduce costs, increase speed, improve customer satisfaction and insight, and build ‘modern’ intelligent applications
What is your use case? What are the capabilities you might need? There’s an AI Service, or a pairing of services that will address the need.
AI Services descriptions for color:
Amazon Rekognition:
Rekognition makes it easy to add image and video analysis to your applications. You just provide an image or video to the Rekognition API, and the service can identify the objects, people, text, scenes, and activities, as well as detect any inappropriate content.
Amazon Rekognition also provides highly accurate facial analysis and facial recognition on images and video that you provide. You can detect, analyze, and compare faces for a wide variety of user verification, people counting, and public safety use cases.Rekognition is a simple and easy to use API that can quickly analyze any image or video file stored in Amazon S3. Amazon Rekognition is always learning from new data, and we are continually adding new labels and facial recognition features to the service.
More info: https://aws.amazon.com/rekognition/
Amazon Polly:
Amazon Polly is a service that turns text into lifelike speech, allowing you to create applications that talk, and build entirely new categories of speech-enabled products.
Polly is a text to speech service that uses advanced deep learning technologies to synthesize speech that sounds like a human voice.
With dozens of lifelike voices across a variety of languages, you can select the ideal voice and build speech-enabled applications that work in many different countries.
More info: https://aws.amazon.com/polly/
Amazon Transcribe:
Amazon Transcribe is an automatic speech recognition (ASR) service that makes it easy for developers to add speech-to-text capability to their applications.
Using the Amazon Transcribe API, you can analyze audio files stored in Amazon S3 and have the service return a text file of the transcribed speech.
Amazon Transcribe can be used for lots of common applications, including the transcription of customer service calls and generating subtitles on audio and video content.
The service can transcribe audio files stored in common formats, like WAV and MP3, with time stamps for every word so that you can easily locate the audio in the original source by searching for the text. Amazon Transcribe is continually learning and improving to keep pace with the evolution of language.
More info: https://aws.amazon.com/transcribe/
Amazon Translate:
Amazon Translate is a neural machine translation service that delivers fast, high-quality, and affordable language translation.
Neural machine translation is a form of language translation automation that uses deep learning models to deliver more accurate and more natural sounding translation than traditional statistical and rule-based translation algorithms.
Amazon Translate allows you to localize content - such as websites and applications - for international users, and to easily translate large volumes of text efficiently.
More info: https://aws.amazon.com/translate/
Amazon Comprehend:
Amazon Comprehend is a natural language processing (NLP) service that uses machine learning to find insights and relationships in text.
The service identifies the language of the text; extracts key phrases, places, people, brands, or events; understands how positive or negative the text is; analyzes text using tokenization and parts of speech; and automatically organizes a collection of text files by topic.
Using these APIs, you can analyze text and apply the results in a wide range of applications including voice of customer analysis, intelligent document search, and content personalization for web applications.
More info: https://aws.amazon.com/comprehend
Amazon Lex:
Amazon Lex is a service for building conversational interfaces into any application using voice and text.
Amazon Lex provides the advanced deep learning functionalities of automatic speech recognition (ASR) for converting speech to text, and natural language understanding (NLU) to recognize the intent of the text, to enable you to build applications with highly engaging user experiences and lifelike conversational interactions.
With Amazon Lex, the same deep learning technologies that power Amazon Alexa are now available to any developer, enabling you to quickly and easily build sophisticated, natural language, conversational bots
More info: https://aws.amazon.com/lex
Easily extract text and data from virtually any document
Amazon Rekognition is a service that applies machine learning to extract information from images and video
You can use the ‘MinConfidence’ parameter in your API requests to balance detection of content (recall) vs the accuracy of detection (precision).
Marinus Analytics has indexed over one and a half million faces in FaceSearch since its inception. The company uses a suite of AWS resources—in addition to Amazon Rekognition—to operate the Traffic Jam platform
Easily extract text and data from virtually any document
detect-syntax, detect-sentiment
Amazon Comprehend helps you comprehend unstructured text by extracting structured information from it.
- extract positive, negative, neutral and mixed sentiment
- extract entities like people, organizations, numbers, dates
- Key phrases gives you the important phrases in the text like “beautiful views” in a hotel review
- English and Spanish for Entities, Sentiment and Keyphrases
- language detection with capability to detect over 100 languages
- The topic modeling API helps you detect topics in a corpus of text. This is an unsupervised algorithm and will work on text in any domain
- We use Deep Learning to power our APIs and this results in higher accuracy and continuous improvement over time with usage.
Syntax:
This API allows customers to tokenize (find word boundaries) words and gives each word its part of speech (noun, pronoun, verb, adjective)
For example, let’s say you’re using the Comprehend Sentiment API and you notice a set of sentences are negative. Using the new Comprehend Syntax API, you can break those sentences into words and their part of speech – from here you can now look the nouns mentioned and the verbs and adjectives used to describe those nouns, allowing you to find out exactly what customers were saying – maybe they didn’t like the color or the price.
Easily extract text and data from virtually any document
We also announced Amazon Polly at last year’s re:Invent. Polly is our text-to-speech service that uses advanced deep learning technologies to synthesize speech that sounds like a human voice.
Key challenges: (1) low latency rendering and delivery, and (2) content ownership (store it on S3 or anywhere, it's your content)
Spoken language crucial for language learning
Accurate pronunciation matters
Faster iteration thanks to TTS
As good as natural human speech
When teaching a foreign language, accurate pronunciation matters. If exposed to incorrect pronunciation, learners develop their listening and speaking skills poorly, which compromises their ability to communicate effectively. Duolingo uses text-to-speech (TTS) to provide high-quality language education. To some, this approach might seem counterintuitive: shouldn’t people learn by listening to a native speaker?
Find a company that records audio in the language: The company must find a voice actor who not only speaks the language, but also who speaks with good pronunciation and clarity.
Find someone to evaluate the quality of pronunciation: We need an independent party from the recording company to create a small sample of sentences, which this party uses to evaluate pronunciation quality of the recordings.
Record and evaluate the quality of the sample sentences.
Set up a contract with the recording company.
Record all sentences.
Evaluate recordings, providing a data quality assurance check. For example, we need to check if all files are in the proper format and correctly separated. This step is necessary because the industry standard is to record all sentences in a single session and separate them later.
Easily extract text and data from virtually any document
Easily extract text and data from virtually any document
Casos de uso: Call centers, Legendas para VOD, Transmissão com closed captions, Transcreva reuniões
It provides timestamps for every word so you can align the text with the audio for subtitling and search use cases
The output text has punctuation, which makes it easy to read
Amazon Transcribe can provide transcription for a wide range of use cases including customer service, subtitling, search, and compliance.
Popular use cases:
Improving Customer Service. By converting audio input into text, Amazon Transcribe lets you build text analytics applications that can search and analyze voice input. Customer contact centers can use Amazon Transcribe to transcribe voice-based interactions, and mine the data for insights using other AWS services like Amazon Comprehend to extract meaning and intent from conversations.
Captioning/Subtitling Workflows. Amazon Transcribe can help content generation and media distributors improve reach and access by automatically generating time-stamped subtitles that can be displayed along with the video content.
Cataloging Audio Archives. The service enables you to transcribe audio and video assets into fully searchable archives for compliance monitoring and risk management. Customers can use Amazon Transcribe to convert audio to text, and use Amazon Elasticsearch service to index and perform text-based search across their audio/video library.
Easily extract text and data from virtually any document
Use cases:
Create smart search indexes
Extract structured data from documents and create a smart index using Amazon Elasticsearch Service to allow you to search through millions of financial statements quickly. For example, a mortgage company could use Amazon Textract to process millions of scanned loan applications in a matter of hours and have the extracted data indexed in Amazon Elasticsearch. This would allow them to create search experiences like “search for loan applications where applicant name is John Doe,” or “search contracts where the interest rate is 2 percent.”
Build automated document processing workflows
Amazon Textract can provide the inputs required to automatically process forms without human intervention. For example, a bank could write code to read PDFs of loan applications. The information contained in the document could be used to initiate all of the necessary background and credit checks to approve the loan so that customers can get instant results of their application rather than having to wait several days for manual review and validation.
Maintain compliance in document archives
Because Amazon Textract identifies data types and form labels automatically, it’s easy to maintain compliance with information controls. For example, an insurer could use Amazon Textract to feed a workflow that automatically redacts personally identifiable information (PII) for their review before archiving claim forms by automatically recognizing the important key-value pairs that require protection.
1/ Excited to introduce Amazon Textract, an OCR++ service to easily extract text and data from virtually any document.
2/ No ML experience required
3/ TRANSITION: Let’s take a look at how this works…
1/ You can try to use previous data to try and make a forecast prediction of what will happen next in a time series; in this example, the sales of a specific product. Today, even when folks try to use a spreadsheet; they often end up basing their forecasts on gut intuition and hope. Not a great strategy.
2/ And there is a real cost to getting these forecasts wrong.
3/ If demand is lower than prediction, you have invested in unsold inventory, which potentially puts the price up of all goods.
4/ If demand is greater than prediction, you leave customers unhappy since your cannot fulfill their demand.
5/ But these forecasts don’t just depend on a single series of data; they rely on potentially millions of seemingly independent variables…
LAUNCH CUSTOMERS: Mercado Libre, CJ Logistics
Customers like MercadoLibre, Latin America's most popular e-commerce site. is using Amazon Forecast to predict demand for over 50,000 different products, Forecast’s state-of-the-art deep learning algorithms available out of the box. Forecast is removing all the heavy lifting of setting up pipelines, re-training schedules, and re-generating forecasts, so they can experiment with hundreds of models very easily.
Under the hood, just like Personalize, Forecast will:
1/ Load the data from your historical time series and related causal data stores, inspecting it, identifying the core features which will build the best model.
2/ Forecast then inspect the data, looking for causals which have a high degree of statistical value, since these will carry the most signal which can be used by the machine learning algorithm
3/ Forecast then selects the best algorithm from the job from the 8 Forecast has under the hood
4/ Prepare the model with an optimal set of configuration options (called hyperparameters), which help the model train more quickly, and identify more ‘signal’ in the data to improve forecast accuracy.
5/ Then Forecast trains the models and optimizes them again for accuracy and performance
6/ Forecast provides fully managed model hosting
7/ Spits out forecasts of next steps in time series you send us
8/ All this data remains yours, and not used for anyone but you. Integrates with KMS to encrypt data.
LAUNCH CUSTOMERS: Domino’s, Navitime, Rbmedia/Recorded Books Inc., Spuul, Zola.
One of our launch customers, Domino’s Pizza, is using Amazon Personalize to predict purchasing behavior and apply context about individual customers and their circumstances to deliver personalized promotions and notifications.
Spuul, a video streaming platform delivering Indian movies and TV shows to an audience worldwide, was manually categorizing and displaying content to users based on broad segmentation. With Amazon Personalize, they now can provide unique and personalized recommendations to each customer.
And Sony Interactive Entertainment (SIE), which is Sony’s video game division overseeing the PlayStation ecosystem, is using Amazon SageMaker and Amazon Personalize to automate and accelerate their machine learning development, and drive more effective personalization at scale.
Another launch customer, Zola, develops innovative wedding planning tools to serve couples. They want to provide the best possible recommendations to our customers based on their style, interests, or preferences. Until now, those recommendations were implemented via rule-based ranking, popularity, or, more recently, via a similarity model calculated offline. With Amazon Personalize, Zola can respond to customer actions in real-time and quickly deliver solutions that would have otherwise taken a much larger team and several months development time.
NAVITIME, a leading provider of navigation technology and services in Japan, is using Amazon Personalize to to improve the accuracy of predictions in their navigation app by personalizing search results as well as recommended navigation routes, based on the users personal preference.
Very excited to announce Amazon Personalize, a real-time personalization and recommendation service, based on the same technology used at Amazon.com - no ML experience required.
What’s happening behind the scenes?
1/ Personalize will load in all of the provided activity streams and data sets; we actually load this into an EMR cluster in the background, ready for analysis.
2/ Personalize then inspects the data, looking for features which have a high degree of statistical value, since these will carry the most signal which can be exploited by the machine learning algorithm.
3/ We look for areas with sparse data and apply what we’ve learned at Amazon to deliver value there
4/ We then select the best algorithms for the job. We can select from up to 6 pre-built models and mix and match if it leads to better results. Personalize will identify multiple algorithms need to be used together to improve accuracy and the personalization experience for end users.
5/ We then prepare the model with an optimal set of configuration options (called hyperparameters), which help the model train more quickly, and identify more ‘signal’ in the data to improve personalization accuracy.
6/ Then we train the models, and optimize them again for accuracy and performance.
7/ Personalize provides fully managed model hosting; along with two components under the hood which enable low latency, real time personalization:
Pre-processed machine learning values in a feature store.
Cached responses to common recommendations.
8/ If this feels a little like the flow of SageMaker, it’s because it’s similar -> difference is Personalize does all this work for you under the covers without your having to take any of the steps yourself
9/ While all these decisions Personalize is making informed by 20+ years of Amazon experience
10/ Data in these resulting models is yours and yours alone
11/ Personalize is like having your own Amazon.com machine learning personalization team at your beck and call, 24 hours a day.
A Pizza Enterprises Ltd (DPE) da Domino é uma das maiores empresas de pizza do mundo; sua visão é ser líder em entregas em todos os bairros. “O cliente está no centro de tudo o que fazemos na Domino's e estamos a trabalhar incansavelmente para melhorar e melhorar a sua experiência. Usando o Amazon Personalize, podemos alcançar a personalização em escala em toda a nossa base de clientes, o que antes era impossível. O Amazon Personalize nos permite aplicar o contexto sobre clientes individuais e suas circunstâncias e fornecer comunicações personalizadas, como ofertas especiais e ofertas por meio de nossos canais digitais.” Mallika Krishnamurthy, Chefe Global, Estratégia & Insights - Domino's Pizza Enterprises