© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Adrian Hornsby, Cloud Architecture Evangelist @ AWS
@adhorn
Moving forward with AI
What is Artificial Intelligence?
“A system or service which can perform tasks
that usually require human intelligence”
Predicting the price of a house with humans
Price
City
ZipCode Life Quality
Parking
Size
# Room
Accessibility
Family Friendly
Predicting the price of a house with neural network
Price
City
ZipCode Life Quality
Parking
Size
# Room
Accessibility
Family Friendly
Input Output
Discovered by the neural network
Artificial Neural Network
The (60 years) rise of Artificial Intelligence
One of the ”Founding Father" of Artificial Intelligence
John McCarthy
1955
Photo from the 1956 Dartmouth
Conference with Marvin Minsky,
Ray Solomonoff, Claude Shannon,
John McCarthy, Trenchard More,
Oliver Selfridge and Nathaniel
Rawchester
Frank Rosenblatt, 1957
Perceptron
First known deep network
https://devblogs.nvidia.com/deep-learning-nutshell-history-training/
Alexey Grigorevich Ivakhnenko, 1965
Paul Werbos, 1975
Backpropagation
LeCun, 1989
First application of
backpropagation
https://www.youtube.com/watch?v=FwFduRA_L6Q
The curse of dimensionality
The Advent of AI
Algorithms
The Advent of AI
Data
Algorithms
The Advent of AI
Data
GPUs
& Acceleration
Algorithms
The Advent of AI
Data
GPUs
& Acceleration
Cloud
Computing
Algorithms
AWS
Common neural networks
& use cases
Convolutional Neural Networks (CNN)
Conv 1 Conv 2 Conv n
…
…
Feature Maps
Labrador
Dog
Beach
Outdoors
Softmax
Probability
Fully
Connected
Layer
https://www.youtube.com/watch?v=qGotULKg8e0
• Over 10 million images from 300,000 hotels
• Using Keras and EC2 GPU instances
• Fine-tuned a pre-trained Convolutional Neural
Network using 100,000 images
• Hotel descriptions now automatically feature the
best available images
CNN: Object Classification
Nuno Castro - Ranking hotel images using deep learning
ImageNet Classification Error Over Time
CNN: Object Detection
https://github.com/precedenceguo/mx-rcnn https://github.com/zhreshold/mxnet-yolo
CNN: Face Detection
https://github.com/tornadomeet/mxnet-face
Autonomous Driving Systems
CNN: Object Segmentation
CNN: Object Segmentation
FDA-approved
medical imaging
https://www.periscope.tv/AWSstartups/1vAGRgevBXRJl
https://www.youtube.com/watch?v=WE81dncwnIc
CNN: Object Segmentation
CNN: Neural Style Transfer
Long Short Term Memory Networks (LSTM)
• LSTM are capable of learning long-term
dependencies
• Designed to recognize patterns in sequences
of data such as:
• text
• genomes
• handwriting
• spoken words
• numerical times series data coming from
sensors, stock markets, etc.
LSTM: Machine Translation
https://github.com/awslabs/sockeye
Generative Adversarial Networks (GAN)
The future at work (already) today
Generating new ”celebrity” faces
https://github.com/tkarras/progressive_growing_of_gans
Generative adversarial networks (GAN)
The future at work (already) today
Semantic labels → Cityscapes street views
https://tcwang0509.github.io/pix2pixHD/
Data, Algorithms, Humans and
Artificial Intelligence
Ground Truth Generation
Training
How much data do you need?
Predicting the price of a house castle
150+ rooms
Rule of thumbs
• Data should cover as many combinations of features as
possible
• More data is almost always better
• Approx. 10x more than the number of features
Most important for you to do today?
“Data is gold”
Pro-tip
• Make it ridiculously easy to collect and store any type of
data.
• One line of code should be all it takes for anyone in the
company to start collecting and storing new data type.
What processes should you boost
with AI?
Where to look at in your organisation ?
• Where data is being analysed to help making decisions.
• Sales
• Marketing
• Social media
• Customer supports 
• Logs
• Etc.
How do you start?
The Low Hanging Fruits
Put AI in the hands of every developer and data scientist
AI @ AWS: Our mission
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Application
Services
Platform
Services
Frameworks
&
Infrastructure
API-driven services: Vision & Language Services, Conversational Chatbots
Deploy machine learning models with high-performance machine learning
algorithms, broad framework support, and one-click training, tuning, and
inference.
Develop sophisticated models with any framework, create managed, auto-
scaling clusters of GPUs for large scale training, or run inference on trained
models.
AI @ AWS – Stack
AI in the hands of every developer and data scientist
Application Services
The low hanging fruits
• API-driven
• Not training required
• Pre-trained on general datasets
• No infrastructure to manage
• Use it now with one line of code
Application
Services
API-driven services: Vision & Language Services, Conversational Chatbots
Amazon
Rekognition
Object and scene detection
Facial analysis
Face comparison
Person Tracking
Celebrity recognition
Image moderation
Text-in-Image
Amazon Rekognition (Image & Video)
Deep learning-based visual analysis service
Marinus Analytics uses facial recognition to
stop human trafficking
“Now with Traffic Jam’s
FaceSearch, powered by
Amazon Rekognition,
investigators are able to
take effective action by
searching through millions
of records in seconds to
find victims.”
http://www.marinusanalytics.com/articles/2017/10/17/amazon-rekognition-helps-marinus-analytics-fight-human-trafficking
Amazon Polly
Hei! Jeg heter Liv.
Skriv inn noe her,
så leser jeg det
opp.
Amazon Polly
Text In, Life-like Speech Out
The Text-To-Speech technology behind Amazon Polly takes advantage of
bidirectional long short-term memory (LSTM)*
* https://www.allthingsdistributed.com/2016/11/amazon-ai-and-alexa-for-all-aws-apps.html
“With Amazon Polly our users benefit from
the most lifelike Text-to-Speech voices
available on the market.”
Severin Hacker
CTO, Duolingo
”
“ Amazon Polly delivers
incredibly lifelike voices
which captivate and engage
our readers.
John Worsfold
Solutions Implementation Manager, RNIB
• RNIB delivers largest library of
audiobooks in the UK for nearly 2
million people with sight loss
• Naturalness of generated speech is
critical to captivate and engage readers
• No restrictions on speech
redistributions enables RNIB to create
and distribute accessible information in
a form of synthesized content
RNIB provides the largest library in the UK for people with sight loss
Amazon Lex
“What’s the weather
forecast?”
“It will be sunny
and 25°C”
Weather
Forecast
Amazon Lex
Build Conversational Chatbots
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
“Hello, what’s up? Do you
want to go see a movie
tonight?”
Amazon Translate
Natural and fluent language translation
"Bonjour, quoi de neuf ? Tu
veux aller voir un film ce
soir ?"
Amazon
Translate
“Hello, this is Allan
speaking”
Amazon Transcribe
Automatic speech recognition service
Amazon
Transcribe
Amazon Comprehend
Discover insights from text
Entities
Key Phrases
Language
Sentiment
Amazon
Comprehend
STORM
WORLD SERIES
STOCK MARKET
WASHINGTON
LIBRARY OF
NEWS ARTICLES *
Amazon
Comprehend
* Integrated with Amazon S3 and AWS Glue
Amazon Comprehend
Support for large data sets and topic modeling
Twitter Stream
API
Kinesis
Lambda
S3 Athena
Translate Comprehend
Transcribe
1 day to build, $17/day to run
(to analyze tweets for
AWS-size customers)
Multilingual Social Analytics
Moving deeper in the rabbit-hole
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Data Visualization &
Analysis
Business Problem –
ML problem framing Data Collection
Data Integration
Data Preparation &
Cleaning
Feature Engineering
Model Training &
Parameter Tuning
Model Evaluation
Are Business
Goals met?
Model Deployment
Monitoring &
Debugging
YesNo
DataAugmentation
Feature
Augmentation
The AI Process
Re-training
Predictions
Say hello to Transfer Learning (hidden gem 1)
• Initialise parameter with pre-trained model
• Use pre-trained model as fixed feature extractor and
build model based on feature
• Why?
• It takes a long time and a lot of resources to train a neural
network from scratch.
Model Zoos (hidden gem 2)
• Full implementations of many state-of-the-art models
reported in the academic literature.
• Complete models, with scripts, pre-trained weights and
instructions on how to build and fine tune these
models.
https://mxnet.apache.org/model_zoo/index.html
End-to-End
Machine Learning
Platform
Zero setup Flexible Model
Training
Pay by the second
$
Amazon SageMaker
Build, train, and deploy machine learning models at scale
Wrapping up
1. Understand what AI is.
2. Take great care of your data.
3. Find the processes that need improvements.
4. Start with the low hanging fruits.
5. Slowly develop yourself into an AI-powered organisation.
There’s Never Been A
Better Time
To Build New Businesses
@adhorn

Moving Forward with AI

  • 1.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved. Adrian Hornsby, Cloud Architecture Evangelist @ AWS @adhorn Moving forward with AI
  • 3.
    What is ArtificialIntelligence?
  • 4.
    “A system orservice which can perform tasks that usually require human intelligence”
  • 5.
    Predicting the priceof a house with humans Price City ZipCode Life Quality Parking Size # Room Accessibility Family Friendly
  • 6.
    Predicting the priceof a house with neural network Price City ZipCode Life Quality Parking Size # Room Accessibility Family Friendly Input Output Discovered by the neural network
  • 7.
  • 8.
    The (60 years)rise of Artificial Intelligence
  • 9.
    One of the”Founding Father" of Artificial Intelligence John McCarthy 1955
  • 10.
    Photo from the1956 Dartmouth Conference with Marvin Minsky, Ray Solomonoff, Claude Shannon, John McCarthy, Trenchard More, Oliver Selfridge and Nathaniel Rawchester
  • 11.
  • 12.
    First known deepnetwork https://devblogs.nvidia.com/deep-learning-nutshell-history-training/ Alexey Grigorevich Ivakhnenko, 1965
  • 13.
  • 14.
    LeCun, 1989 First applicationof backpropagation https://www.youtube.com/watch?v=FwFduRA_L6Q
  • 15.
    The curse ofdimensionality
  • 16.
    The Advent ofAI Algorithms
  • 17.
    The Advent ofAI Data Algorithms
  • 18.
    The Advent ofAI Data GPUs & Acceleration Algorithms
  • 19.
    The Advent ofAI Data GPUs & Acceleration Cloud Computing Algorithms AWS
  • 24.
  • 26.
    Convolutional Neural Networks(CNN) Conv 1 Conv 2 Conv n … … Feature Maps Labrador Dog Beach Outdoors Softmax Probability Fully Connected Layer
  • 27.
    https://www.youtube.com/watch?v=qGotULKg8e0 • Over 10million images from 300,000 hotels • Using Keras and EC2 GPU instances • Fine-tuned a pre-trained Convolutional Neural Network using 100,000 images • Hotel descriptions now automatically feature the best available images CNN: Object Classification Nuno Castro - Ranking hotel images using deep learning
  • 28.
  • 29.
  • 30.
  • 31.
  • 32.
  • 33.
  • 34.
  • 35.
    Long Short TermMemory Networks (LSTM) • LSTM are capable of learning long-term dependencies • Designed to recognize patterns in sequences of data such as: • text • genomes • handwriting • spoken words • numerical times series data coming from sensors, stock markets, etc.
  • 36.
  • 37.
    Generative Adversarial Networks(GAN) The future at work (already) today Generating new ”celebrity” faces https://github.com/tkarras/progressive_growing_of_gans
  • 38.
    Generative adversarial networks(GAN) The future at work (already) today Semantic labels → Cityscapes street views https://tcwang0509.github.io/pix2pixHD/
  • 39.
    Data, Algorithms, Humansand Artificial Intelligence
  • 40.
  • 41.
    How much datado you need?
  • 42.
    Predicting the priceof a house castle 150+ rooms
  • 43.
    Rule of thumbs •Data should cover as many combinations of features as possible • More data is almost always better • Approx. 10x more than the number of features
  • 44.
    Most important foryou to do today? “Data is gold”
  • 45.
    Pro-tip • Make itridiculously easy to collect and store any type of data. • One line of code should be all it takes for anyone in the company to start collecting and storing new data type.
  • 46.
    What processes shouldyou boost with AI?
  • 47.
    Where to lookat in your organisation ? • Where data is being analysed to help making decisions. • Sales • Marketing • Social media • Customer supports  • Logs • Etc.
  • 48.
    How do youstart? The Low Hanging Fruits
  • 49.
    Put AI inthe hands of every developer and data scientist AI @ AWS: Our mission
  • 52.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved.
  • 53.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved.
  • 54.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved.
  • 55.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved.
  • 56.
    Application Services Platform Services Frameworks & Infrastructure API-driven services: Vision& Language Services, Conversational Chatbots Deploy machine learning models with high-performance machine learning algorithms, broad framework support, and one-click training, tuning, and inference. Develop sophisticated models with any framework, create managed, auto- scaling clusters of GPUs for large scale training, or run inference on trained models. AI @ AWS – Stack AI in the hands of every developer and data scientist
  • 57.
    Application Services The lowhanging fruits • API-driven • Not training required • Pre-trained on general datasets • No infrastructure to manage • Use it now with one line of code Application Services API-driven services: Vision & Language Services, Conversational Chatbots
  • 58.
    Amazon Rekognition Object and scenedetection Facial analysis Face comparison Person Tracking Celebrity recognition Image moderation Text-in-Image Amazon Rekognition (Image & Video) Deep learning-based visual analysis service
  • 60.
    Marinus Analytics usesfacial recognition to stop human trafficking “Now with Traffic Jam’s FaceSearch, powered by Amazon Rekognition, investigators are able to take effective action by searching through millions of records in seconds to find victims.” http://www.marinusanalytics.com/articles/2017/10/17/amazon-rekognition-helps-marinus-analytics-fight-human-trafficking
  • 61.
    Amazon Polly Hei! Jegheter Liv. Skriv inn noe her, så leser jeg det opp. Amazon Polly Text In, Life-like Speech Out The Text-To-Speech technology behind Amazon Polly takes advantage of bidirectional long short-term memory (LSTM)* * https://www.allthingsdistributed.com/2016/11/amazon-ai-and-alexa-for-all-aws-apps.html
  • 62.
    “With Amazon Pollyour users benefit from the most lifelike Text-to-Speech voices available on the market.” Severin Hacker CTO, Duolingo
  • 63.
    ” “ Amazon Pollydelivers incredibly lifelike voices which captivate and engage our readers. John Worsfold Solutions Implementation Manager, RNIB • RNIB delivers largest library of audiobooks in the UK for nearly 2 million people with sight loss • Naturalness of generated speech is critical to captivate and engage readers • No restrictions on speech redistributions enables RNIB to create and distribute accessible information in a form of synthesized content RNIB provides the largest library in the UK for people with sight loss
  • 65.
    Amazon Lex “What’s theweather forecast?” “It will be sunny and 25°C” Weather Forecast Amazon Lex Build Conversational Chatbots
  • 66.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved.
  • 67.
    “Hello, what’s up?Do you want to go see a movie tonight?” Amazon Translate Natural and fluent language translation "Bonjour, quoi de neuf ? Tu veux aller voir un film ce soir ?" Amazon Translate
  • 68.
    “Hello, this isAllan speaking” Amazon Transcribe Automatic speech recognition service Amazon Transcribe
  • 69.
    Amazon Comprehend Discover insightsfrom text Entities Key Phrases Language Sentiment Amazon Comprehend
  • 70.
    STORM WORLD SERIES STOCK MARKET WASHINGTON LIBRARYOF NEWS ARTICLES * Amazon Comprehend * Integrated with Amazon S3 and AWS Glue Amazon Comprehend Support for large data sets and topic modeling
  • 71.
    Twitter Stream API Kinesis Lambda S3 Athena TranslateComprehend Transcribe 1 day to build, $17/day to run (to analyze tweets for AWS-size customers) Multilingual Social Analytics
  • 72.
    Moving deeper inthe rabbit-hole
  • 73.
    © 2017, AmazonWeb Services, Inc. or its Affiliates. All rights reserved. Data Visualization & Analysis Business Problem – ML problem framing Data Collection Data Integration Data Preparation & Cleaning Feature Engineering Model Training & Parameter Tuning Model Evaluation Are Business Goals met? Model Deployment Monitoring & Debugging YesNo DataAugmentation Feature Augmentation The AI Process Re-training Predictions
  • 74.
    Say hello toTransfer Learning (hidden gem 1) • Initialise parameter with pre-trained model • Use pre-trained model as fixed feature extractor and build model based on feature • Why? • It takes a long time and a lot of resources to train a neural network from scratch.
  • 75.
    Model Zoos (hiddengem 2) • Full implementations of many state-of-the-art models reported in the academic literature. • Complete models, with scripts, pre-trained weights and instructions on how to build and fine tune these models.
  • 76.
  • 77.
    End-to-End Machine Learning Platform Zero setupFlexible Model Training Pay by the second $ Amazon SageMaker Build, train, and deploy machine learning models at scale
  • 78.
  • 79.
    1. Understand whatAI is. 2. Take great care of your data. 3. Find the processes that need improvements. 4. Start with the low hanging fruits. 5. Slowly develop yourself into an AI-powered organisation.
  • 80.
    There’s Never BeenA Better Time To Build New Businesses @adhorn