Our mission is to take our rich experience and expertise with machine learning across Amazon and put it in the hands of all organizations--every developer, data scientist, reseacher. Said another way, we want to simplify machine learning. We want to make it easy for all developers to easily build intelligent applications
Broadest and deepest set of AI and ML services for your business Some organizations are product focused (‘build like we build!’), some are competitor focused (‘fast follow what’s successful’). At Amazon we start with the customer need, and work backwards from there; 90% of the AWS roadmap is built directly from customer feedback; the other 10% is where customers don’t quite know what to ask for, and so we read between the lines on their behalf. We launched over 200 new machine learning features and services in the last year; no other provider has done more than half as many. Developers wake up and have material new capabilities available at their fingertips virtually every day. And our pace is only accelerating And we are unmatched in the breadth of ML services we offer to customers today—capabilities to serve the diverse use cases and the needs of all ML practitioners, at all skill levels.
Solving the toughest problems holding back ML: cost, ease of use, and data Cost – Machine learning training and inference can be expensive. With AWS, you can train and deploy ML models on the highest-performing lowest-cost compute environment with our on-demand compute model. As well, organizations can take advantage of the free tier to get started with AI Services East of use – The machine learning workflow can be difficult, time consuming and iterative, which leaves many organizations and developers thinking machine learning is complex and difficult to use. There are many steps involved from prepping data, choosing algorithms, building, training and deploying models…and iterating over and over again. With AWS, AI Services are ready-made, no machine learning skills required. We’ve developed and pre-trained these services so you can easily add intelligence to your applications. And with Amazon SageMaker, we’ve simplified the entire workflow to build, train, optimize, and deploy machine learning models. Data – Customers are continuously aggregating data from various sources but they don’t necessarily know how to derive value from all that data. Whether it’s figuring out the right data sources and types (unstructured vs structured, proprietary vs public), and how do they store and manage all this data. AWS provides a comprehensive data lake, data storage and analytics to prepare and analyze data essential for machine learning.
Unmatched support for the most popular frameworks We believe customer choice and flexibility are incredibly important. For example, when it comes to frameworks, some companies are focused on pushing customers to just one framework, but we have a different perspective. It is way to early to be locked into any single framework, particularly when they all have different strengths and developers have different preferences We're not going to try to tell you that you should try to solve all your machine learning and deep learning problems with one framework. We support all the popular frameworks and interfaces to give you choice and flexibility.
Built on the most comprehensive cloud platform optimized for Machine Learning In order to do machine learning successfully, you not only need machine learning capabilities, but a comprehensive platform--You need the right data store, security, and analytics services to work together. With AWS, you can choose from a comprehensive set of services for data analysis including data warehousing, business intelligence, batch processing, stream processing, data workflow orchestration. We also have the deepest set of security and encryption features, with the broadest accreditation Our approach to research is fundamentally different: Instead of abstract (and large, expensive, disconnected) R&D teams in an ivory tower, our scientists are embedded with our product and engineering teams. They work on advancing research, but do so in close proximity to our customers – working backwards from their needs closely aligned with our engineering teams (so they can pick up their research and bring it to customers more quickly). We invent and simplify on behalf of our customers in a way few other organizations do. It’s this unique approach that fuels our pace of innovation
Most of this is being done on AWS, where we have tens of thousands of customers who are running machine learning on AWS with twice as many references or customers here than you'll find anywhere else. It's a broad and fast growing group that includes NBC Universal, Snap, State Farm Insurance, Intuit, F1 Racing, NFL, Coinbase, Convoy, and Sony Interactive.
NBC Universal NBCUniversal is one of the world’s leading media companies in the development, production and marketing of entertainment, news and information to a global audience. 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. They require frame accurate detection of content events, as well as creating metadata to support compliance, marketing and advertising, as well accessibility and censorship responsibilities. NBCU is investing in the rich capabilities that Amazon Rekognition delivers to assist and automate many of these manual processes.
FINRA 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. 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. 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. This exponentially increases the scope of investigations while shortening the time to conclusion.
Sky News Sky News is a 24-hour international multimedia news organization based in the UK. They wanted to identify celebrity guests at the Royal Wedding in real time, to improve the experience for viewers of their broadcast. They used the GrayMeta Platform powered by Amazon Rekognition to deliver the “Who’s Who Live” feature in the broadcast. This provided viewers with biographical information of identified guests as they arrived at the event.
Kia Kia Motors, the oldest automobile manufacturer in South Korea, builds more than three million vehicles a year for customers in 180 countries. They use Amazon Rekognition for advanced image and video analysis of an in-car camera that detects the driver. Then, the car automatically adjusts driver-assistance features like personalized mirror and seat positioning.
State Farm: State Farm is a family of insurance and financial services companies that together serve tens of millions of customers in the U.S. Motor vehicle accident claims can take longer when a vehicle is sent to repair, but is later determined to be a total loss. When automating this decision point, rules-based models struggled to make that determination with a high level of precisions. State Farm is creating new machine learning models that can better predict if a vehicle should be designated a total loss. This can streamline the claims process, improving the experience for their customers.
Intuit Intuit is a provider of accounting and financial management solutions including TurboTax and QuickBooks. Intuit’s small business and sole proprietor customers, such as ride share drivers, struggle to itemize all of their qualifying business expenses. Using Amazon SageMaker, Intuit developed ML models that can pull a year’s worth of bank transactions then find deductible business expenses for customers. These models can identify thousands of dollars in additional deductions for Intuit’s customers. And with SageMaker, Intuit has reduced machine learning deployment time from 6 months to 1 week.
F1 Formula 1 is the highest class of automobile racing sanctioned by the FIA, with the fastest and most advanced cars. During each race, 120 sensors on each car generate 3 GB of data with more than 1,500 data points being created each second. Using Amazon SageMaker, Formula 1’s data scientists are training deep-learning models with 65 years of historical race data to extract critical race performance statistics, make race predictions, and give fans insight into the split-second decisions and strategies adopted by teams and drivers. By streaming real-time race data to AWS using Amazon Kinesis, Formula 1 can capture and process key performance data for each car during every twist and turn of the Formula 1 circuits. Then, by deploying advanced machine learning via Amazon SageMaker, Formula 1 can pinpoint how a driver is performing and whether or not drivers have pushed themselves over the limit. By sharing these insights through television broadcasts and digital platforms, Formula 1 allows fans access to the inner workings of their favorite teams and drivers.
MLB Major League Baseball (MLB) is the oldest of the four major professional sports leagues in the United States and Canada. MLB has been using AWS to collect and distribute game-day stats to enhance the fan experience since 2015, including data from the pitcher, catcher, and runner on stolen base attempts. MLB and the Amazon ML Solutions Lab trained a deep neural network to predict stolen base success by using numerous data including runner’s speed and burst, catcher’s pop time, pitcher’s velocity and handedness, lead-off distance, and the game situation. The model was quickly deployed using Amazon SageMaker, providing sub-second response times required for integrating predictions into in-game graphics in real time, and on ML instances that auto-scale across multiple availability zones. This improves the in-game experience for MLB fans watching on television.
NFL The NFL governs and oversees America’s most popular spectator sport, American football. More than 90% of NFL fans are catching the games on TV, online, and mobile, and the NFL is constantly searching for ways to improve the fan experience. Built using Amazon SageMaker, NextGen Stats enables the NFL to predict formations, play outcomes, routes, and key events in a game. The data is used to enhance broadcasts during games and provide insights for post-game analysis.
Convoy Convoy is a Seattle-based logistics company that is disrupting the $700 billion trucking industry. Because trucking in the US is made up of fragmented shippers and haulers that align through human brokers, 40 percent of truck miles are done with an empty truck. Using Amazon SageMaker, Convoy’s machine learning models analyze millions of shipping jobs and truckers, and then recommend matches that are cost-efficient and timely for both shippers and truckers. This helps fill otherwise empty trucks, increasing their profitability while at the same time reducing vehicle emissions.
Coinbase Coinbase is a digital wallet and exchange platform where over 20 million merchants and consumers have traded more than $150 billion in cryptocurrencies. To combat fraud, Coinbase needed a scalable solution to verify uploaded images of IDs. The created a machine learning model that compares an uploaded ID with millions of others in their database. Since online fraudsters often use the same photo for multiple IDs, this helps them identify potentially fraudulent accounts. This lets Coinbase on-board customers faster, and lets them provide low risk customers with higher purchase limits and fewer restrictions.
Sony Interactive (PlayStation) Sony Interactive Entertainment (SIE) is Sony’s video game division, overseeing the PlayStation ecosystem. SIE uses Amazon SageMaker to create intelligent interfaces that connect PlayStation users with the right content or people at the right time. This includes a recommendation engine to display suggested games and content for purchase in the PlayStation store. Using SageMaker, SIE modernized the Playstation Store, using predictive ML to drive highly personalized customer experiences, improve enterprise data reporting and drive product feature innovation.
Bill.com Bill.com is the leading digital business payments company, providing AP and AR solutions. As they enter new markets, they need an efficient way to comply with anti-money-laundering and anti-fraud regulations. They implemented a fraud detection model using Amazon SageMaker, creating a scalable and automated approach to compliance in new markets. They anticipate this functionality will allow them to serve new markets and customers around the world.
Chick-fil-A Chick-fil-A, Inc. is a family owned and privately held restaurant company that operates more than 2,300 restaurants. They needed a way to monitor the freshness of their fries so their staff can be alerted when a new batch is to be made. Chick-fil-A uses MXNet on SageMaker to train computer vision models that identify fried food which needs to be replaced. The solution alerts workers to start preparing a new batch, helping ensure friend food meets their rigorous quality standards. At this point, they have built a successful prototype that they are now working to move to production to increase their guest experience. They anticipate this will increase the quality of food delivered to their customers.
NuData – MasterCard NuData Security, A MasterCard Company, is trusted by some of the largest global brands to verify online users. Fraudsters are increasingly using sophisticated automation and other techniques to circumvent even multi-factor authentication. 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. This means their approach to fraud detection is continually adapting, changing, and evolving to keep pace with fraudsters.
Siemens Siemens Financial Services (SFS) finances infrastructure, equipment, and working capital investments, and acts as a manager of financial risks within Siemens AG. When conducting due diligence on energy projects, SFS personnel would need to manually review thousands of pages of financial documents in order to extract the relevant information. SFS used TensorFlow on Amazon SageMaker to develop a natural language processing (NLP) model that accelerates due diligence by extracting the most relevant and critical information from supporting documents. With the help of the model, SFS has reduced time to summarize diligence documents from hours down to 30 seconds. Beth Isreal Deaconess Medical Center Beth Isreal Deaconess Medical Center (BIDMC) is the teaching hospital of Harvard Medical School located in the heart of Boston. Before commencing surgery, staff needed to verify the patient had completed the proper consent form. However, this paper form has no standard format and was often buried in scanned copies of the patient’s medical record. This was delaying or cancelling hundreds of surgeries a year. BIDMC uses TensorFlow on AWS to automatically review faxed medical records for each patient, identifying and sharing the consent form with hospital staff. This helps BIDMC to better utilize their human and physical resources, while removing unneeded expense due to delays and cancellations.
Snap Snap are the makers of Snapchat, a camera and sharing app with more than 180 million active users. They need to serve relevant ads to their users, in order to maintain user engagement and driver return for their advertisers. Snap created a customer deep learning solution using TensorFlow models on Amazon SageMaker to improve how they match ads to users. They anticipate this solution will increase user engagement with advertising. They also expect using SageMaker will allow their data scientists to spend less time on training models and more time developing new AI/ML solutions.
At AWS, we think about machine learning as having three macro layers in the stack. The bottom layer is for expert machine learning practitioners. These are people who are comfortable designing their own tools and workflows to build, train, tune, and deploy models. They're comfortable operating at the framework and infrastructure level.
The middle layer of the stack is for ML developers and data scientists where we’ve simplified the entire ML workflow to make it easier for them to build, train and deploy models.
The top layer is a set of AI services that should allow you to be even more effective in your everyday activities in building a better customer experience and a better business for your customers.
This is the second year in a row where you've seen this plethora of services that we've launched for builders at every layer of that stack. It's never been easier and faster and more cost effective for everyday builders to build, train, tune, and deploy machine learning models than it is today. As we've said, it's not just machine learning models and services that allow you to do machine learning at scale the way you need to.
They're really useful and there's a huge number of these at AWS at this point, but to really do machine learning the right way, it starts with having the right secure, operationally performant, fully featured, cost-effective data lake or data store with the right access control on some of your most valuable data and the broadest array of analytic services and that broadest array of machine learning services at every layer of the stack.
There's nobody across those dimensions that has a set of capabilities like AWS, which is why the vast majority of companies are using AWS for machine learning.
AWS Philosophy: give you great primitives to build with, plus abstractions to make common tasks really easy as well
let’s talk about the top layer of the stack, AI Services:
Let's talk about the top layer of the machine learning stack. We talked about the bottom layer for expert machine learning practitioners and the middle layer for everyday developers and data scientists. This top layer is for companies and builders who don't want to mess with the models at all. They just want to plug into built models effectively through an API. And we have built a number of these services over the last couple years that we've delivered for you. If you want to know, “Here's an object, what's in the object?” Or “Here's a video, tell me what's in the video.” Or "Is this a face?" Or "Does this face match a set of faces that I (this customer) have given you as a set of faces?" Those are all part of our computer vision service called Amazon Rekognition. Or some customers say, "Here's text. Turn it into speech." And that's what we built Polly for. Or they say, "Here's all this audio. Transcribe it to text." And that's why we built Amazon Transcribe. "Here's this transcribed text. Translate it into lots of languages." That's what Amazon Translate does. "Here are these corpuses of transcribed, translated texts. Tell me what the heck is in it so I don't have to look or have humans do it." And that's what you use natural language processing or Amazon Comprehend for. It’s a broad set of these top layer services that mimic human cognition, what people often call AI. But if you think about it, so much of the world's data is still locked away in documents.
Amazon Polly is our text-to-speech service that uses advanced deep learning technologies to synthesize speech that sounds like a human voice.
Works for a wide range of audio quality. Specially trained to work well with 8Khz telephony audio Integrated with S3 so you can point to a file in S3 and it will transcribe it for you 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
You can transcribe your contact center calls and analyze the text with natural language processing to find insights into customer behavior. You can use transcribe to generate subtitles for your video content. You can also use this text to show your viewers the right advertisements based on the content You can generate closed captions for increased accessibility of your content You can transcribe meetings so that you can search through your archives for relevant information
Amazon Translate is a Neural Machine Translation Service on AWS that offers fast and affordable text translation for any multilingual application
Over the last year we productized proprietary state-of-the-art Neural Machine Translation that uses Deep Learning models to perform language to language transformation in context, producing fluent and complete sentences.
Neural MT was a leap ahead in terms of current performance as well as the future potential.
Let me explain. In essence statistical models were built by taking millions of sentences, breaking them up into smaller phrases and then using a fancy lookup algorithm to create new sentences based on probabilities that the phrase is more commonly used. The main problem with these models was that they didn’t understand context, and they were limited to about five words around the one they were translating.
That’s no longer the case with neural engines. I am extremely over simplifying this, but generally speaking these engines are inspired by the way that the human brain learns and processes information. They understand context, focus, and recognize named entities. And this makes a huge difference.
In terms of speed, Neural Machine Translation (NMT) is ~ x300 faster than a human and ~ x1,500 more cost efficient These numbers are calculated for a normal 8-hour work day but of course, machines can work 24x7 increasing productivity to x4 that AND new instances of service can be spawned, unlike humans…
The paradigm shift became possible with the breakthrough in QUALITY. As of today, on average, NMT achieves 85% of human parity, meaning NMT is ~85% as accurate as a professional translator.
Epsilon research indicates 80% of consumers are more likely to make a purchase when brands offer personalized experiences: http://pressroom.epsilon.com/new-epsilon-research-indicates-80-of-consumers-are-more-likely-to-make-a-purchase-when-brands-offer-personalized-experiences/
If you think of language as a form of personalizing customer experiences than it makes sense that Common Sense Advisory finds that HALF of consumers AND businesses prefer content in their own language even if it is less than perfect. They want familiar experiences and language is a key aspect of that.
In fact, it is so important that MORE THAN HALF of consumers make language a pivotal decision criteria regarding their purchase habits. If they cannot find it, research it or get post-sale service in their own language, they will NOT buy it…
Amazon Translate is a general-purpose neural machine translation that offers Translation between 21 languages and a total 417 combinations Covers 85% of Economic Activity @
It costs $15/1M characters. So that news article I mentioned before would down to around 7.5 cents…@
It can support real time applications with a latency of under 500 ms for the average sentence and 150 ms for an average short message like an Alexa utterance. @
Sentence tag handling ensures that source to target text styling and formatting is maintained. Useful for translating HTML files. @
Data security - Your data is protected by AWS’ security standards. You can use SSL certificates to encrypt the data in transit, use IAM to control access to your resources, and rest assured that your data is stored securely. @
Easy to use – Getting started with Amazon Translate is easy. It is available through multiple SDKs, and as we saw, most implementations require THREE lines of code @
Amazon Comprehend is a natural language processing service that uses machine learning to find insights and relationships in text. Amazon Comprehend identifies the language of the text; extracts key phrases, places, people, brands, or events; understands how positive or negative the text is; and automatically organizes a collection of text files by topic.
Our customers are using Amazon Comprehend to identify key topics, entities, and sentiments in social media and news streams, and to enhance their ability to access and aggregate unstructured data from the vast document libraries that exist within their organizations.
Hotels.com has thousands of customer views and comments that are submitted by people who stay at the properties. It’s historically been difficult to find what matters in all this data. By using Amazon Comprehend, Hotels.com is able to uncover the unique characteristics that people like or don’t like about each hotel. Consequently, the company is better able to make recommendations to their users.
Available in English, Spanish, French, German, Italian, and Portuguese
Use Cases: Content Personalization: Customers are using Amazon Comprehend NLP output to identify related documents based on entities, phrases or even topic similarities to drive content personalization and recommendations Semantic Search: Customers are using Amazon Comprehend to index entities for boosting search results.
Intelligent data warehouse: Customers are using Amazon Comprehend to convert unstructured data in relational databases into structured data and then inserting it back into the data warehouse so that it can be queried
Intelligent data warehouse: Customers are using Amazon Comprehend to convert unstructured data in relational databases into structured data and then inserting it back into the data warehouse so that it can be queried
Social Analytics: Customers are using Amazon Comprehend to ingest, process and analyze trends from social media posts across Twitter and Facebook
Information management: Customers are using Amazon Comprehend for indexing and finding related content for enterprise information management and various internal business processes including compliance and IT
Amazon Comprehend helps you comprehend unstructured text by extracting structured information from it.
Amazon Comprehend can extract positive, negative, neutral and mixed sentiment It can extract entities like people, organizations, numbers, dates Key phrases gives you the important phrases in the text like “beautiful views” in a hotel review We support English and Spanish for Entities, Sentiment and Keyphrases We have 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.
Amazon Lex is our service that enables customers to build conversational interfaces using voice and text.
Accurate time series forecasting service, based on the same technology used at Amazon.com. No machine learning experience required. Draws from 20 years of experience in forecasting at Amazon
Packages a suite of 8 algorithms that includes 5 deep-learning algorithms and 3 statistical methods. The deep-learning algorithms improve accuracy by up to 50%, for datasets with over 1000 time-series.