This document provides an overview of Amazon's artificial intelligence capabilities including services like Amazon Polly for text-to-speech, Amazon Rekognition for image and video analysis, and Amazon Lex for conversational interfaces. It discusses how these services use deep learning techniques and frameworks like Apache MXNet. Case studies are presented showing how companies like Duolingo and HubSpot utilize Amazon AI services to improve their products.
3. Artificial Intelligence at Amazon
Thousands Of Employees Across The Company Focused on AI
Discovery &
Search
Bring Machine
Learning To All
Fulfilment &
Logistics
Enhance
Existing Products
Define New
Categories Of
Products
8. The Advent Of
Deep Learning
Data
GPUs
& Acceleration
Programming
models
Algorithms
9. The Advent Of
Deep Learning
Data
GPUs
& Acceleration
Programming
models
Algorithms
AWS
10. AI Services
AI Platform
AI Engines
Amazon
Rekognition
Amazon
Polly
Amazon
Lex
More to come
in 2017
Amazon
Machine Learning
Amazon EMR
(MapReduce)
Spark &
SparkML
More to come
in 2017
Apache
MXNet
Caffe Theano KerasTorch CNTKTensorFlow
P2 and G2
GPUs
ECS Lambda
AWS
Greengrass
FPGAEC2 CPU
More to
come
in 2017
Hardware
Introducing Amazon AI
14. Programmable Portable High Performance
Near linear scaling
across hundreds of GPUs
Highly efficient
models for mobile
and IoT
Simple syntax,
multiple languages
aws.amazon.com/mxnet
Apache MXNet
15. Most Open Best On AWS
Optimized for
deep learning on AWS
Accepted into the
Apache Incubator
(Integration with AWS)
Why Apache MXNet?
16. Apache MXNet is the deep learning framework
of choice for Amazon
17. Deep Learning Framework Comparison
Apache MXNet TensorFlow Cognitive Toolkit
Industry Owner
N/A – Apache
Community
Google Microsoft
Programmability
Imperative and
Declarative
Declarative only Declarative only
Language
Support
R, Python, Scala,
Julia, C++, Javascript,
Go, Matlab and more..
Python, C++,
Experimental Go and
Java
Python, C++ and
Brainscript
Code Length|
AlexNet (Python)
44 sloc 107 sloc using TF.Slim 214 sloc
Memory Footprint
(LSTM)
2.6GB 7.2GB N/A
18. P2 INSTANCES DEEP TEMPLATEDEEP LEARNING AMI
github.com/awslabs/deeplearning-cfn
Getting Started with MXNet
20. Converts text
to life-like speech
47 voices 24 languages Low latency,
real time
Fully managed
Amazon Polly: life-like Speech Service
aws.amazon.com/polly
22. “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
23. 2. Intelligible and Easy to Understand
1. Automatic, Accurate Text Processing
Polly: A Focus On Voice Quality & Pronunciation
A measure of how comprehensible speech is.
”Peter Piper picked a peck of pickled peppers.”
24. 2. Intelligible and Easy to Understand
3. Add Semantic Meaning to Text
“Richard’s number is 2122341237“
“Richard’s number is 2122341237“
Telephone Number
1. Automatic, Accurate Text Processing
Polly: A Focus On Voice Quality & Pronunciation
25. 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.”
1. Automatic, Accurate Text Processing
Polly: A Focus On Voice Quality & Pronunciation
26. Duolingo voices its language learning service Using Polly
Duolingo is a free language learning service where
users help translate the web and rate translations.
With Amazon Polly our users
benefit from the most lifelike
Text-to-Speech voices
available on the market.
Severin Hacker
CTO, Duolingo
”
“ • Spoken language crucial for
language learning
• Accurate pronunciation matters
• Faster iteration thanks to TTS
• As good as natural human speech
34. Bynder allows you to easily create, find and use
content for branding automation and marketing
solutions.
With our new AI capabilities,
Bynder’s software… now allows
users to save hours of admin
labor when uploading and
organizing their files, adding
exponentially more value.
Chris Hall
CEO, Bynder
”
“
With Rekognition, Bynder revolutionizes marketing admin tasks with AI capabilities
40. Voice & Text
“Chatbots”
Powers
Alexa
Voice interactions
on mobile, web
& devices
Text interaction
with Slack and FB Messenger
Enterprise
Connectors
Salesforce
Microsoft Dynamics
Marketo
Zendesk
Quickbooks
Hubspot
Lex: Build Natural, Conversational Interactions in
Voice & Text
aws.amazon.com/lex
41. HubSpot is an inbound marketing and sales platform
that helps companies attract visitors, convert leads,
and close customers.
Through Amazon's Lex, we're adding
sophisticated natural language
processing capabilities that helps
GrowthBot provide a more intuitive UI
for our users. Amazon Lex lets us take
advantage of advanced A.I. and
machine learning without having to
code the algorithms ourselves.
Dharmesh Shah
Founder, HubSpot
”
“
With Lex, HubSpot adds conversational interfaces
HubSpot's GrowthBot is an all-in-one chatbot
which helps marketers and sales people be
more productive by providing access to
relevant data and services using a
conversational interface. With GrowthBot,
marketers can get help creating content,
researching competitors, and monitoring their
analytics.
46. USE CASE
‣ Build the infrastructure:
‣ Collect sensors and user data
‣ Execute ML algorithms to predict tyre wear
‣ Scale and extend to other predictive service
48. ML CURRENT IMPLEMENTATION
‣ Tyre wear algorithm
‣ We wanted to use Lambda functions but we are falling back on dockerized micro
services and SQS. This mainly because:
‣ execution time limitation
‣ problems in installing Python dependency
‣ Rancher is in charge of machines spinning for scaling computation
49. WHY AWS?
‣ Already the cloud provider company-wide
‣ Authentication (kinda) out of the box with Cognito
‣ Data Lake built on S3 + EMR gives more flexibility in terms of:
‣ Region support / deployment
‣ Scalability (switching on and off transient EMR clusters)
‣ Overall TCO
‣ Parallelize ML algos execution through containers (EC2 used as repo)
‣ Support for ML services evolutions:
‣ Spark on EMR
‣ Kinesis for streaming
50. FUTURE STEPS
‣ Additional algorithms / services
‣ Working on streaming data (or at least nearer to real time) and testing Kinesis
‣ Have on line models evaluation and selections through Spark and transient EMR
clusters
‣ Possibility of using Amazon managed services for ML
‣ Build robust alerts and sanity check data pipeline