Machine Learning is having a major impact in our society, but how can we simplify the build, train, and deploy process for all developers and data scientists? Understand how cloud-based machine learning frameworks can help turn your data into intelligence. We will introduce the general machine learning process utilising the AWS Deep Learning AMIs and hear from carsales.com.au about how they built the Cyclops, a Super Human Image Recognition Software on AWS. We will then discuss the new capabilities delivered by Amazon SageMaker and how this product will further reduce the undifferentiated heavy lifting; freeing you up to focus on your business and allow your developers to quickly and easily expand into the world of Machine Learning.
6. Types of Machine Learning
Supervised
Develop a predictive model based on both input and output data.
E.g. Photo classification or tagging
Unsupervised
Discover an internal structure based on input data only
E.g. Auto-classification of documents based on context
7. How is deep learning different?
Deep Learning
AlgorithmData
Automatically
identifies features!!!
Model
Unseen Sample
Prediction
Deep Learning
Statistical
Machine Learning
Feature
engineeringData Model
Unseen Sample
Prediction
Features
Shallow Learning
8. Haven’t I Heard All This Before? Why Now?
AlgorithmsVast data Compute
14. Rekognition Images
Use Cases:
Facial recognition - Missing persons, Wanted persons
Image analysis – Object and Scene detection
Intelligent image tagging - Including celebrity recognition
Content filtering - Adult content detection
Extract text from images - License plates, Street names etc.
15. Rekognition Video
Rekognition Video with a little processing: [(14, 18), (26, 40), (48, 57), (69, 88)]
Given a photo of Andy Jassy Tell me where you find him in this video…
16. Rekognition Video
Rekognition Video with a little processing: [(14, 18), (26, 40), (48, 57), (69, 88)]
Given a photo of Andy Jassy Tell me where you find him in this video…
18. Use Cases:
Facial recognition - Missing persons, Wanted persons
Video analysis – object, scene and activity detection
Intelligent video tagging (including celebrity recognition)
Content filtering (Unsafe content detection)
Automatic highlight extraction
Rekognition Video
19. Comprehend
I love these shoes. Because of back issues, I am unable to
wear shoes with any kind of heel but these are so cute that
I decided to give them a try. They are character (ballroom)
shoes so the arch support is spot on. The fit is amazing and
the quality is better than I expected for the price
Given this review Tell me what the sentiment is…
Comprehend:
Sentiment: POSITIVE 0.99
20. No stars... This show is a JOKE! When I received the shoes I
was so excited! In the picture it's a cute retro little pump. In
person it's awful... A full size too large and absurdly
wide.the materials are so cheap I've seen better in
children's costume shoes!
Comprehend
Given this review Tell me what the sentiment is…
Comprehend:
Sentiment: NEGATIVE 0.74
21. text = "I love these shoes. Because of back issues, I am unable to wear shoes with
any kind of heel but these are so cute that I decided to give them a try. They are
character (ballroom) shoes so the arch support is spot on. The fit is amazing and
the quality is better than I expected for the price"
comprehend_client = boto3.client( 'comprehend', 'us-west-2’ )
result = comprehend_client.detect_dominant_language( Text = text )
dominant_language = result['Languages'][0]['LanguageCode’]
result = comprehend_client.detect_sentiment( Text = text ,
LanguageCode = dominant_language )
sentiment = result['Sentiment']
Comprehend
22. Comprehend
Use Cases:
Monitor sentiment - Twitter, Facebook, Blog, Comment etc.
Identify the language of text
Automated content tagging
Document analysis - enable rich text search
23. Transcribe
Ladies and gentlemen, please welcome chief executive officer of
amazon web services andy jassy. Thank you and welcome to the
sixth annual a ws reinvent this is our very favorite time of the
year it's our favorite week and we're really excited that you're
spending the week with us. You're here with forty three
thousand of your peers
Given this video or audio clip create a transcription…
25. Transcribe
Use Cases:
Transcription of customer service calls
Automatic minute taking
Generate subtitles on audio and video content
Enable search of video and audio
33. Fetch data
Clean &
format data
Prepare &
transform
data
Train
model
Evaluate
model
The Machine Learning Process Is Hard …
34. Fetch data
Clean &
format data
Prepare &
transform
data
Train
model
Evaluate
model
Integration
with prod
Monitor /
debug /
refresh
The Machine Learning Process Is Hard …
35. Fetch data
Clean &
format data
Prepare &
transform
data
Train
model
Evaluate
model
Integration
with prod
Monitor /
debug /
refresh
6-18
months
The Machine Learning Process Is Hard …
36. A fully managed service that enables data scientists and developers to
quickly and easily build machine-learning based models into production smart applications.
Amazon SageMaker
Now in Sydney!
38. Build, train, and deploy machine learning models at scale
End-to-End
Machine Learning
Platform
Zero setup Flexible Model
Training
Pay by the
second
Amazon SageMaker
40. XGBoost, FM,
Linear Learner,
DeepAR for
forecasting
and
classification
Kmeans, PCA,
and BlazingText
(Word2Vec) for
clustering,
dimensionality
reduction and
pre-processing
Image
classification
with
convolutional
neural networks
LDA and
NTM for
topic
modeling,
seq2seq for
translation
Random Cut
Forest for
anomaly
detection
SageMaker Built-in Algorithms
66. Linear Learner
In this example, each feature was a pixel grey-scale
value between 0-1
Imagine using this for housing loan valuations.
Each feature is a feature of the property
67. Call To Action #1
What business problem can ML help you solve?
FRAUD
68. • Getting started with Amazon SageMaker:
https://aws.amazon.com/sagemaker/
• Use the Amazon SageMaker SDK:
• For Python: https://github.com/aws/sagemaker-python-sdk
• For Spark: https://github.com/aws/sagemaker-spark
• SageMaker Examples: https://github.com/awslabs/amazon-
sagemaker-examples
• Let us know what you build!
Call To Action #2