3. A Flywheel For Data
More Data Better Analytics
Better Products
4. A Flywheel For Data
More Data Better Analytics
Better ProductsMore Users
5. A Flywheel For Data
More Data Better Analytics
Better ProductsMore Users
Click stream
User activity
Generated content
Purchases
Clicks
Likes
Sensor data
6. A Flywheel For Data
More Data Better Analytics
Better ProductsMore Users
Click stream
User activity
Generated content
Purchases
Clicks
Likes
Sensor data
Object Storage
Databases
Data warehouse
Streaming analytics
BI
Hadoop
Spark/Presto
Elasticsearch
7. A Flywheel For Data
More Data Better Analytics
Better ProductsMore Users
Click stream
User activity
Generated content
Purchases
Clicks
Likes
Sensor data
Object Storage
Databases
Data warehouse
Streaming analytics
BI
Hadoop
Spark/Presto
Elasticsearch
Artificial
Intelligence
8. A Flywheel For Data
More Data Better Analytics
Better ProductsMore Users
Click stream
User activity
Generated content
Purchases
Clicks
Likes
Sensor data
Object Storage
Databases
Data warehouse
Streaming analytics
BI
Hadoop
Spark/Presto
Elasticsearch
Artificial
Intelligence
9. • Artificial Intelligence: design software applications which
exhibit human-like behavior, e.g. speech, natural language
processing, reasoning or intuition
• Machine Learning: teach machines to learn without being
explicitly programmed
• Deep Learning: using neural networks, teach machines to
learn from complex data where features cannot be explicitly
expressed
11. Artificial Intelligence At Amazon
Thousands Of Employees Across The Company Focused on AI
Discovery &
Search
Fulfilment &
Logistics
Enhance
Existing Products
Define New
Categories Of
Products
Bring Machine
Learning To All
13. Machine Learning for customer support
Not all customer interactions can be solved in a self-service mode.
Therefore, Amazon operates large customer support centers where
Customer Service Representatives (CSR) handle customer requests.
The machine learning models described above are used to optimize the
human interactions of these requests.
For example, they are used to route the customer call to the best CSR
before the customer has even started to speak! They are also used
again during the call.
22. The Advent Of
Deep Learning
Data
GPUs
& Acceleration
Programming
models
Algorithms
23. Can We Help Customers
Put Intelligence At The Heart Of
Every Application & Business?
24. Questions, questions…
What’s the business problem my IT has failed to solve
Should I design and train my own Deep Learning model?
Should I use a pre-trained model?
Should I use a SaaS solution?
Same questions as “Big Data” years ago
26. Amazon AI: Three New Deep Learning Services
Polly
Life-like Speech
27. Amazon AI: Three New Deep Learning Services
Rekognition
Life-like Speech Image Analysis
Polly
28. Amazon AI: Three New Deep Learning Services
Rekognition Lex
Life-like Speech Image Analysis Conversational
Engine
Polly
29. Amazon AI: Three New Deep Learning Services
Polly Rekognition Lex
Life-like Speech Image Analysis Conversational
Engine
30. Polly: Life-like Speech Service
Converts text
to life-like speech
48 voices 24 languages Low latency,
real time
Fully managed
31. “Today in Seattle, WA, it’s 11°F”
‘"We live for the music" live from the Madison Square Garden.’
1. Automatic, Accurate Text Processing
Polly: A Focus On Voice Quality & Pronunciation
32. Polly: A Focus On Voice Quality & Pronunciation
2. Intelligible and Easy to
Understand
1. Automatic, Accurate Text Processing
33. 2. Intelligible and Easy to Understand
3. Add Semantic Meaning to Text
“Richard’s number is 2122341237“
“Richard’s number is 2122341237“
Telephone Number
Polly: A Focus On Voice Quality & Pronunciation
1. Automatic, Accurate Text Processing
34. 2. Intelligible and Easy to Understand
3. Add Semantic Meaning to Text
4. Customized Pronunciation
“My daughter’s name is Kaja.”
“My daughter’s name is Kaja.”
Polly: A Focus On Voice Quality & Pronunciation
1. Automatic, Accurate Text Processing
35. Polly: Life-like Speech Service
High quality,
through
best-in-class
deep learning
Deep
functionality
Easy to use
& thoughtfully integrated
Built for
production
Low
cost
36. Amazon AI: Three New Deep Learning Services
Polly Rekognition Lex
Life-like Speech Image Analysis Conversational
Engine
45. Amazon AI: Three New Deep Learning Services
Polly Rekognition Lex
Life-like Speech Image Analysis Conversational
Engine
46. Lex: Build Natural, Conversational Interactions In Voice & Text
Voice & Text
“Chatbots”
Powers
Alexa
Voice interactions
on mobile, web
& devices
Text interaction
with Slack & Messenger
Enterprise
Connectors
(with more coming) Salesforce
Microsoft Dynamics
Marketo
Zendesk
Quickbooks
Hubspot
50. Origin
Destination
Departure Date
Flight Booking
“Book a flight to
London”
Automatic
Speech Recognition
Natural Language
Understanding
Book Flight
London
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
London Heathrow
51. Origin
Destination London Heathrow
Departure Date
Flight Booking
“Book a flight to
London”
Automatic
Speech Recognition
Natural Language
Understanding
Book Flight
London
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
London Heathrow
52. Origin Seattle
Destination London Heathrow
Departure Date
Flight Booking
“Book a flight to
London”
Automatic
Speech Recognition
Natural Language
Understanding
Book Flight
London
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
London Heathrow
LocationLocation
53. Origin Seattle
Destination London Heathrow
Departure Date
Flight Booking
“Book a flight to
London”
Automatic
Speech Recognition
Natural Language
Understanding
Book Flight
London
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
London Heathrow
Prompt
LocationLocation
“When would you like to fly?”
54. Origin Seattle
Destination London Heathrow
Departure Date
Flight Booking
“Book a flight to
London”
Automatic
Speech Recognition
Natural Language
Understanding
Book Flight
London
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
London Heathrow
Prompt
LocationLocation
“When would you like to fly?”
“When would you like to
fly?”
Polly
57. Origin Seattle
Destination London Heathrow
Departure Date 11/18/2016
Flight Booking
“Next Friday”
Automatic
Speech Recognition
Natural Language
Understanding
Next Friday
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
11/18/2016
58. Origin Seattle
Destination London Heathrow
Departure Date 11/18/2016
Flight Booking
“Next Friday”
Automatic
Speech Recognition
Natural Language
Understanding
Next Friday
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
11/18/2016
59. Origin Seattle
Destination London Heathrow
Departure Date 11/18/2016
Flight Booking
“Next Friday”
Automatic
Speech Recognition
Natural Language
Understanding
Next Friday
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
11/18/2016
Confirmation
“Your flight is booked for next Friday”
60. Origin Seattle
Destination London Heathrow
Departure Date 11/18/2016
Flight Booking
“Next Friday”
Automatic
Speech Recognition
Natural Language
Understanding
Next Friday
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
11/18/2016
“Your flight is booked for
next Friday”
Confirmation
“Your flight is booked for next Friday”
Polly
61. Origin Seattle
Destination London Heathrow
Departure Date 11/18/2016
Flight Booking
“Next Friday”
Automatic
Speech Recognition
Natural Language
Understanding
Next Friday
Grammar
Graph
Utterances
Knowledge
Graph
Flight booking
11/18/2016
Hotel Booking
64. Amazon Machine Learning
• Easy-to-use, managed machine learning service built for
developers
• Robust, powerful machine learning technology based on
Amazon’s
internal systems
• Create prediction and classification models using your data
already
stored in the AWS Cloud
• Deploy models to production in seconds
65. Fully managed model and prediction services
End-to-end service, with no servers to provision and
manage
One-click production model deployment
Programmatically query model metadata to enable
automatic retraining workflows
Monitor prediction usage patterns with Amazon
CloudWatch metrics
66. ”
“
Fraud.net Uses AWS to Quickly, Easily Detect Online Fraud
Fraud.net is the world’s leading crowdsourced
fraud prevention platform.
Amazon Machine Learning
helps us reduce complexity and
make sense of emerging fraud
patterns.
• Needed to build and train a larger number of
more targeted machine-learning models
• Uses Amazon Machine Learning to provide
more than 20 models
• Easily builds and trains models to effectively
detect online payment fraud
• Reduces complexity and makes sense of
emerging fraud patterns
• Saves clients $1 million weekly by helping
them detect and prevent fraud
Oliver Clark
CTO,
Fraud.net
”
“
67. ”
“
Upserve Uses AWS to Help Restaurants Predict Business
Upserve provides online payment and analytical
software to thousands of restaurant owners
throughout the U.S.
Using Amazon Machine
Learning, we can predict the
total number of customers
who will walk through a
restaurant’s doors in a night.
• Needed its restaurant management
platform to provide more predictive
analytics
• Builds and trains more than 100
machine learning models weekly
• Streams restaurant sales and menu
item data in real time
• Helps restaurateurs predict nightly
business
Bright Fulton
Director of Infrastructure Engineering,
”
“
69. Apache MXNet: Open Source library for Deep Learning
Programmable Portable High Performance
Near linear scaling
across hundreds of GPUs
Highly efficient
models for mobile
and IoT
Simple syntax,
multiple languages
Most Open Best On AWS
Optimized for
Deep Learning on AWS
Accepted into the
Apache Incubator
More information at mxnet.io
70. Input Output
1 1 1
1 0 1
0 0 0
3
mx. sym. Convol ut i on( dat a, ker nel =( 5, 5) , num_f i l t er =20)
mx. sym. Pool i ng( dat a, pool _t ype=" max" , ker nel =( 2, 2) ,
st r i de=( 2, 2)
l st m. l st m_unr ol l ( num_l st m_l ayer , seq_l en, l en, num_hi dden, num_embed)
4 2
2 0
4=Max
1
3
...
4
0.2
-0.1
...
0.7
mx. sym. Ful l yConnect ed( dat a, num_hi dden=128)
2
mx. symbol . Embeddi ng( dat a, i nput _di m, out put _di m = k)
0.2
-0.1
...
0.7
Queen
4 2
2 0
2=Avg
Input Weights
cos(w, queen ) = cos(w, k i n g) - cos(w, m an ) + cos(w, w om an )
mx. sym. Act i vat i on( dat a, act _t ype=" xxxx" )
" r el u"
" t anh"
" si gmoi d"
" sof t r el u"
Neural Art
Face Search
Image Segmentation
Image Caption
“ People Riding
Bikes”
Bicycle, People,
Road, Sport
Image Labels
Image
Video
Speech
Text
“ People Riding
Bikes”
Machine Translation
“ Οι άνθρωποι
ιππασίας ποδήλατα”
Events
mx. model . FeedFor war d model . f i t
mx. sym. Sof t maxOut put
72. One-Click GPU
Deep Learning
AWS Deep Learning AMI
Up to~40k CUDA cores
Apache MXNet
TensorFlow
Theano
Caffe
Torch
Pre-configured CUDA drivers
Anaconda, Python3
+ CloudFormation template
+ Container Image
Using AWS, C-SPAN can sample a frame every six seconds for recognition against indexed faces in a database of 97,000 people. Previously, this was done manually: Indexers scrolled through screen captures to identify who was speaking at any given point and select an image to represent each individual in each video. C-SPAN expects to save 8,000 to 9,000 hours a year in labor by automating that process using Rekognition, and will be able to index 100% of its incoming footage and archives.
Today, we have announced Amazon ML, the newest addition to the Amazon Web Services family.
Amazon ML is easy to use, and intended for developers – people who are already most connected and familiar with data instrumentation, pipelines and storage/
Amazon ML is based on the same robust ML technology that is already used within Amazon’s internal systems, generating billions of predictions weekly
Amazon ML is built to make it simple and reliable to use the data that you are already storing in the AWS cloud, in products like Amazon S3, Amazon Redshift and Amazon RD
And lastly, Amazon ML is built to eliminate the gap between having models and using these models to build smart applications. Production deployment is only a click away – and sometimes you won’t even need that one click.
I want to start by talking about ease of use, because it was so incredibly important to us as we built this product. We wanted to make machine learning accessible to all AWS developers. What we know from both our internal experience and speaking with our customers is that using machine learning to build real-world falls into two types. The first stage is experimentation: trying out different data sources, different ways to transform data, different ways to use ML technology to approach a problem. In this stage, it is really helpful to work with interactive tools that let you do many experiments easily.
The second stage is productizing the models. This involves automating model training, deployment and use – for example, tying the prediction generation activities with your application’s business logic. And there is just no substitute for having a capable API to achieve this – that enables the ML piece to act just as robustly as the rest of your application.
Most ML toolkits out there do one or the other. We knew that to enable AWS developers to easily use ML we needed to support both types of ML development. And so what you will see in Amazon ML is a powerful service console that enables you to connect to your data source, to train new models, to measure their quality, and to manage these models easily, without writing any code. This enables you to spend more time thinking about your business problem and your data, not tinkering with the technology pieces. Once you have models that are ready to be productized, you have a full API available to deploy these models, to query them for predictions, and to train and evaluate new models on schedule. It’s a complete lifecycle. We have included Amazon ML API support into all the major AWS SDKs. And, we have also bundled prediction querying capability into the AWS mobile SDKs, making it really simple to create mobile applications that use predictions.
Next, let’s talk about technology.
Amazon ML is based on the same ML technology that has long been deployed within Amazon, and is used to generate tens of billions of predictions weekly.
When I say technology, I mean not just the learning algorithm – which is important but by no means only part of ML systems. Amazon Machine Learning comes with technology that suggests data transformations based on your data that will improve model’s quality – and you are able to use these transformations as they are, or adjust the transformation instructions without writing any data transforming code. Amazon ML also includes functionality to provide alerts when known pitfalls are encountered in your input data – for example, when some attributes have many missing values, or when the model was evaluated with data that is significantly different from what was used to train it. This – ensuring that data is evaluated on a fair dataset is an example of an industry best practice that we have built into the product, among many others, all around the goal of making the resulting models more powerful.
Finally, a word about speed and scale. Amazon ML can create models from up to 100 GB of data. You can use it to generate billions of predictions, and obtain them in batches or real-time. I will get to the prediction interfaces soon.
STORY BACKGROUND
Fraud.net uses Amazon Machine Learning to support its machine-learning models.
The company uses Amazon DynamoDB and AWS Lambda to run code without provisioning and managing servers.
Uses Amazon Redshift for data analysis.
SOLUTION & BENEFITS
Launches and trains machine-learning models in almost half the time it took on other platforms.
Reduces complexity and makes sense of emerging fraud patterns.
Saves customers $1 million each week.
CONTENT TAGS
Main use case: Big Data, Analytics, & Business Intelligence (BI)
Keywords: online fraud, fraud detection, machine learning, big data, Amazon Machine Learning, business intelligence
AWS Services used: Amazon DynamoDB, Amazon Redshift, Amazon Machine Learning, Amazon S3, AWS Lambda
Benefits Realized: Agility, Better Performance, Ease of Use, Lower Cost, Reliability, Scalability/Elasticity, Speed
STORY BACKGROUND
Provider of cloud-based restaurant management platform
Wanted to help restaurant owners gain predictive analytics
Sought to take advantage of machine learning technology
SOLUTION & BENEFITS
Uses Amazon Machine Learning to provide predictive analysis through its Shift Prep application
Builds and trains more than 100 machine learning models each week
Helps restaurant owners predict nightly business
CONTENT TAGS
Main use case: Big Data, Analytics, & Business Intelligence (BI)
Keywords: restaurants, predictive analysis, machine learning, Amazon Machine Learning, online payment process
AWS Services used: Amazon EC2 Container Service, Amazon Data Pipeline, Amazon Machine Learning, Amazon S3, Amazon RDS, Amazon DynamoDB, Amazon Elastic Map Reduce
Benefits Realized: Ease of Use, Lower Time-to-Market, Reliability, Scalability/Elasticity, Speed, User Experience