• Overview of AI services • Deploy a data science environment with the AWS Deep Learning AMI • Leverage the power of Intel GPU’s with NVIDIA Tesla M60 GPUs • Train and deploy Deep Learning models at scale

1. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.

2. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Artificial Intelligence on AWS
Alex Coqueiro
Solutions Architecture Team
Public Sector Canada, Latin America & Caribbean
September 20th, 2017
@alexbcbr

3. technology which can perform tasks that
usually require human intelligence
Artificial Intelligence

5. Running AI in Production on AWS

6. How does it work ?
Historical Data Model Building Prediction
What is my color?
And what is mine?

7. … but more data …
Less Data More Data Even More Data
Linear Models
Categorical Models
Bayesian Methods
Decision Trees
Neural Networks
Cluster Analysis
… More …
Kernel Based Methods

8. Intelligent Services Powered By Deep Learning
Amazon AI

9. AI Services
AI Platform
AI Engines
Amazon
Rekognition
Amazon
Polly
Amazon
Lex
Amazon
Machine Learning
Amazon Elastic
MapReduce
Spark &
SparkML
Caffe Theano KerasTorch CNTK
Amazon AI: Democratized AI
TensorFlow
P2 ECS Lambda
AWS
Greengrass
FPGAEMR/Spark InfrastructureG2/G3
Apache
MXNet

10. …Business Problem…
1
2

11. …Business Problem…
1
2

12. Amazon Rekognition
Deep learning-based image recognition service
Search, verify, and organize millions of images
Object and Scene
Detection
Facial
Analysis
Face
Comparison
Facial
Recognition

13. Face Comparison
Measure the likelihood that faces in two images are of
the same person
• Add face verification to
applications and devices
• Extend physical security
controls
• Provide guest access to
VIP-only facilities
• Verify users for online
exams and polls

14. Amazon Rekognition API
var compareParams = {
SimilarityThreshold: 90,
SourceImage: { ... },
TargetImage: { ... }
}};
rekognition.compareFaces(compareParams, function(err, data) {
if (err) {
console.log(err, err.stack); // an error occurred
} else {
if (data.FaceMatches.length > 0) {
//get item in dynamo
console.log("Similarity: " + data.FaceMatches[0].Similarity);
dynamodb.getItem(paramsItem, function(err, data) { ... }
}
});

15. …Business Problem…
1
2

17. Deep Learning – Neural Network
 Output
Neural Network
Input
Hidden layers
Computing systems inspired by the biological neural networks
which learn to do tasks by considering examples, generally
without task-specific programming

18. Deep Learning – Neural Network

20. AWS Deep Learning AMI
Up to 40,000
CUDA Cores
Frameworks &
Libraries
Python Notebooks
& Examples
https://aws.amazon.com/amazon-ai/amis/

21. AWS
Instance
Type
High
Memory
X1
Compute
Optimized
C4
Storage
Optimized
D2
General
Purpose
M4
FPGA
F1
GPU
Optimized
P2
Graphics
Optimized
G2
Graphics
Optimized
G3
Intel Processor
Intel Xeon E7-
8880 v3
Custom Intel
Xeon E5-2666 v3
Custom Intel
Xeon E5-2676 v3
Custom Intel
Xeon E5-2676 v3
Intel Xeon
E5-2686 v4
Intel Xeon
E5-2686 v4
Intel Xeon
E5-2670
Intel Xeon E5-
2686 v4
Intel AVX AVX 2.0 AVX 2.0 AVX 2.0 AVX 2.0 Yes Yes Yes Yes
Intel AES-NI Yes Yes Yes Yes Yes Yes Yes Yes
Intel Turbo Boost Yes Yes Yes Yes Yes Yes Yes Yes
Network
Performance
25 Gbps 10 Gbps 10 Gbps 25 Gbps 25 Gbps 25 Gbps 10 Gbps 25 Gbps
SSD Storage
EBS Optimized
by default
EBS Optimized
by default
No
EBS Optimized
by default
Yes EBS only Yes EBS only
Amazon EC2 Instances with Intel® Technologies

22. Apache MXNet
Programmable Portable High Performance
Near linear scaling
across hundreds of GPUs
Highly efficient
models for mobile
and IoT
Simple syntax,
multiple languages

23. Frameworks de Deep Learning

24. IMPERATIVE
NDARRAY API
DECLARATIVE
SYMBOLIC
EXECUTOR
>>> import mxnet as mx
>>> a = mx.nd.zeros((100, 50))
>>> b = mx.nd.ones((100, 50))
>>> c = a + b
>>> c += 1
>>> print(c)
>>> import mxnet as mx
>>> net = mx.symbol.Variable('data')
>>> net = mx.symbol.FullyConnected(data=net, num_hidden=128)
>>> net = mx.symbol.SoftmaxOutput(data=net)
>>> texec = mx.module.Module(net)
>>> texec.forward(data=c)
>>> texec.backward()
Mixed Programming Paradigm

25. …Business Problem…
1
2

26. Amazon Polly: Life-like Speech Service
Converts text
to life-like
speech
47 voices 24 languages Low latency,
real time
Fully managed

27. TEXT
Market grew by > 20%.
WORDSPHONEMES
{
{
{
{
{
ˈtwɛn.ti
pɚ.ˈsɛnt
ˈmɑɹ.kət ˈgɹu baɪ ˈmoʊɹ
ˈðæn
PROSODY CONTOURUNIT SELECTION AND ADAPTATION
TEXT PROCESSING
PROSODY MODIFICATIONSTREAMING
Market grew by more
than
twenty
percent
Speech units
inventory

28. Markup language: SSML
Speech Synthesis Markup Language
is a W3C recommendation, an XML-based markup language for speech
synthesis applications
<speak>
My name is Coqueiro. It is spelled
<prosody rate='x-slow'>
<say-as interpret-as="characters">Coqueiro</say-as>
</prosody>
</speak>

29. Converting text to life-like speech
# Invoke Polly client
polly = Session().client("polly")
#Voice synthesize
response = polly.synthesize_speech(
Text=data,
TextType="ssml",
OutputFormat="mp3",
VoiceId="Vitoria”
)

30. Invoking MXNet models
# Library Importing
import mxnet as mx
import numpy as np
# Model loading
sym, arg_params, aux_params = load_model(f_symbol_file.name, f_params_file.name)
# Parameters
mod = mx.mod.Module(symbol=sym)
mod.bind(for_training=False, data_shapes=[('data', (1,3,224,224))])
mod.set_params(parameters, auxiliar_parameters)
# Model execution
labels = predict(url, mod, synsets)
https://aws.amazon.com/blogs/compute/seamlessly-scale-predictions-with-aws-lambda-and-mxnet/

31. Summary
Amazon
S3
Camera
Amazon
DynamoDB
Amazon
EC2 MXNet
Amazon
Rekognition
AWS Lambda
Amazon
EchoAmazon
S3
MRI
Historial Data
Amazon
Polly
MRI Scan

32. AI Services
AI Platform
AI Engines
Amazon
Rekognition
Amazon
Polly
Amazon
Lex
Amazon
Machine Learning
Amazon Elastic
MapReduce
Spark &
SparkML
Caffe Theano KerasTorch CNTK
Amazon AI: Democratized AI
TensorFlow
P2 ECS Lambda
AWS
Greengrass
FPGAEMR/Spark InfrastructureG2/G3
Apache
MXNet

33. Thank You !
Alex Coqueiro
Solutions Architecture Team
Public Sector Canada, Latin America & Caribbean
September 20th, 2017
@alexbcbr