3. Democratizing AI and Machine Learning
Three flavors of Machine Learning
Machine Learning APIs
Building custom ML solutions on Google Cloud
Customer Success Stories
Agenda
7. Two Models Of Computation
Turing Machines
Von-Neumann Architecture
Algorithms Programmed by Humans
Symbolic Vector Space
Brain Machines
Biologically Inspired (Evolutionarily Evolved) Architecture
Algorithms Learned from Experience
8. Confidential & Proprietary
Source: Data scientists= Kaggle Data scientist community , Developers: Evans Data Corporation the figure in 2016 was 21m
State of the Industry: Lack of Expertise
Very few users today
can create a custom ML model.
To democratize AI,
we need to make AI accessible
to millions more
1000’s
Deep Learning
Researchers
21M
Developers
Confidential & Proprietary
<1M
Data Scientists
9. Confidential & ProprietaryConfidential & Proprietary
UPDATEDEPLOYEVALUATETUNE ML MODEL
PARAMETERS
ML MODEL DESIGN
DATA
PREPROCESSING
State of the Industry: Complex & Time Intensive
Large computational resource . Machine learning expertise . Manual data labeling
11. Confidential & ProprietaryGoogle Cloud Platform 11
Rapidly accelerating use of deep learning at Google
Google Cloud Platform 11
Google directories containing Brain Models
2012 2013 2014 2015
3000
2000
1000
0
Used across products:
4000
2016
Uniqueprojectdirectories
2017
14. Proprietary + Confidential
ML Frameworks:
Total Control for Power
Users
Machine Learning APIs:
Ready to Go
Cloud
Vision API
Cloud
Translation API
Cloud Natural
Language API
Cloud
Speech API
Cloud ML EngineTensorFlow
Cloud
Jobs API
Cloud Video
Intelligence API
AutoML:
Bring Your Own Data
(We Do the Rest)
Pick Your Flavor
Cloud DataprocSpark ML
16. Proprietary + Confidential
ML Frameworks:
Total Control for Power
Users
Machine Learning APIs:
Ready to Go
Cloud
Vision API
Cloud
Translation API
Cloud Natural
Language API
Cloud
Speech API
Cloud ML EngineTensorFlow
Cloud
Jobs API
Cloud Video
Intelligence API
AutoML:
Bring Your Own Data
(We Do the Rest)
Pick Your Flavor
Cloud DataprocSpark ML
18. Vision API
Detect broad sets of
categories within an image,
ranging from modes of
transportation to animals.
Analyze facial features to
detect emotions: joy,
sorrow, anger.
Detect logos.
Detect and extract text
within an image, with
support for a broad range of
languages, along with
support for automatic
language identification.
Extract text
Detect different types of
inappropriate content from
adult to violent content.
Powered by Google Safe
Search
Detect inappropriate contentObject Recognition Facial sentiment & logos
TRY THE API
19. Natural Language API
Identify entities and label by
types such as person,
organization, location,
events, products and media.
Enables you to easily
analyze text in multiple
languages including
English, Spanish and
Japanese.
Extract tokens and
sentences, identify parts of
speech (PoS) and create
dependency parse trees for
each sentence.
Syntax analysisEntity Recognition Multi-Language Support
TRY THE API
Understand the overall
sentiment expressed in a
block of text.
Sentiment Analysis
20. Speech API
Powered by deep
learning neural
networking to power
your applications..
No need for signal
processing or noise
cancellation before
calling API. Can
handle noisy audio
from a variety of
environments.
Noise Robustness
Can provide context
hints for improved
accuracy. Especially
useful for device and
app use cases.
Word HintsSpeech Recognition
TRY THE API
Recognizes over 80
languages & variants.
Can also filter
inappropriate content
in text results
Over 80 languages
Can stream text
results, returning
partial recognition
results as they
become available.
Can also be run on
buffered or archived
audio files.
Real-time results
21. Translation API
Supports more than 100
languages and thousands
of language pairs.
Behind the scenes,
Translation API is learning
from logs analysis and
human translation
examples. Existing
language pairs improve and
new language pairs come
online at no additional cost.
Sometimes you don’t know
your source text language in
advance. Can automatically
identify languages with high
accuracy.
Automatic language
detection
The Premium edition is
tailored for users who need
precise, long-form
translation services (e.g.
livestream translations, high
volume of emails, detailed
articles and documents)
Premium edition BETA
Text Translation Continuous Updates
TRY THE API
22. Video Intelligence API
Detect entities within the
video, such as "dog",
"flower" or "car".
You can now search your
video catalog the same way
you search text
documents..
Extract actionable insights
from video files without
requiring any machine
learning or computer vision
knowledge.
Enable Video Search
More features will be added
to the Video Intelligence API
during the BETA period.
More to come ... BETA
Label Detection Insights From Videos
27. Proprietary + Confidential
ML Frameworks:
Total Control for Power
Users
Machine Learning APIs:
Ready to Go
Cloud
Vision API
Cloud
Translation API
Cloud Natural
Language API
Cloud
Speech API
Cloud ML EngineTensorFlow
Cloud
Jobs API
Cloud Video
Intelligence API
AutoML:
Bring Your Own Data
(We Do the Rest)
Pick Your Flavor
Cloud DataprocSpark ML
29. Confidential & ProprietaryGoogle Cloud Platform 29
AI expertise + Data
+ Computation
We call this
AutoML
Current solution But we can turn this into
Data + 100x Computation
30. How does AutoML work?
Controller: proposes ML models Train & evaluate models
20K
times
Iterate to
find the
most
accurate
model
31. Confidential & ProprietaryGoogle Cloud Platform 31
AI does AI
Systematic exploration
of the model space, using
the techniques finessed in
AlphaGo, yields super-human
performance in AI network design
33. AutoML for Cloud Customers
Dataset Baseline AutoML
Customer 1 (Media) 75% 99%
Customer 2 (Housing) 87% 93%
Customer 3 (Wildlife) 85% 95%
Customer 4 (Sports) 90% 96%
Customer 5 (Insurance) 0.7 mean AUC 0.95 mean AUC
Results for AutoML on image problems https://cloud.google.com/automl/
34. Proprietary + Confidential
ML Frameworks:
Total Control for Power
Users
Machine Learning APIs:
Ready to Go
Cloud
Vision API
Cloud
Translation API
Cloud Natural
Language API
Cloud
Speech API
Cloud ML EngineTensorFlow
Cloud
Jobs API
Cloud Video
Intelligence API
AutoML:
Bring Your Own Data
(We Do the Rest)
Pick Your Flavor
Cloud DataprocSpark ML
36. ● PaaS for Tensorflow
● Instantly scale your training up to 100 workers
(industry leading)
● Automatic monitoring and logging
● Seamlessly transition from training
to prediction
● Built in model version management
● No lock-in. Option to download your trained
models for on-premise or mobile deployment
Cloud ML Engine
37. Automatically tune your model with HyperTune
● Automatic hyperparameter tuning service
● Build better performing models faster and save
many hours of manual tuning
● Google-developed search algorithm efficiently
finds better hyperparameters for your
model/dataset
● Flexible: Hyperparameters are provided to to
user code as command line flags, allowing any
post-processing you want.
HyperParam #1
Objective
Want to find this
Not these
HyperParam
#2
39. Google Use Of TensorFlow: # of Models
Search
Gmail
Translate
Maps
Android
Photos
Speech
YouTube
Play
… many others ...
Production use in many areas:
Research use for:
100s of projects and papers
Internal TensorFlow launch
40.
41. Google-designed custom ASIC built
and optimized for TensorFlow
1st generation used in production for
over 16 months
Now on 2nd generation—180 Teraflops
per TPU
TensorFlow Research Cloud—1000
TPUs for researchers, at no charge.
Tensor Processing Unit
42. TPU Pod
64 2nd-gen TPUs
11.5 petaflops
4 terabytes of memory
2-D toroidal mesh network
44. Proprietary + Confidential
Define ML use cases
Define specific ML use cases
for the project
Select algorithm
Choose the right ML
algorithm for the task
Build ML model
Develop the first iteration
of the ML model
Present results
Present results of the model in
a way that demonstrates its
value to stakeholders
Iterate ML model
Refine the ML model to
improve performance and
efficacy
Data pipeline &
feature engineering
Create the right features
from raw data for the
ML task
Plan for deployment
Prepare for deployment in
production
Operationalize model
Deploy and operationalize
ML model in production
Monitor model
Monitor deployed ML model
and retrain or rebuild when
performance degrades
1 3
10 789
Data exploration
Perform exploratory data
analysis to understand the
data
2 4
6
5
Start
a new ML project
with PSO
Cloud Discover Cloud MVM Cloud Deploy
Machine Learning Lifecycle
45. Building an ML model requires 3 things
Data
Compute
Talent
Data Scientist
Software Engineer
52. Talent
● Every organization has people
capable of building ML
systems
● But those people may not
have the training and tools
they need to be successful
with machine learning
● Google provides both
53. Training
● Google Professional Services
will bring Google Machine
Learning expertise to your
company
● Intensive trainings and
workshops from 1 to 4 weeks
● Customized to your needs
Talent
54. Proprietary + Confidential
We Can Help You Implement your Solution
Google PSO1 Google + Partner2 3
Implement
Machine
Learning
APIs
Build
ML Models
Cloud
ML Engine
Deploy and ManageAnalyze and Plan
Google Cloud
Google Cloud
55. Proprietary + Confidential
Define ML use cases
Define specific ML use cases
for the project
Select algorithm
Choose the right ML
algorithm for the task
Build ML model
Develop the first iteration
of the ML model
Present results
Present results of the model in
a way that demonstrates its
value to stakeholders
Iterate ML model
Refine the ML model to
improve performance and
efficacy
Data pipeline &
feature engineering
Create the right features
from raw data for the
ML task
Plan for deployment
Prepare for deployment in
production
Operationalize model
Deploy and operationalize
ML model in production
Monitor model
Monitor deployed ML model
and retrain or rebuild when
performance degrades
1 3
10 789
Data exploration
Perform exploratory data
analysis to understand the
data
2 4
6
5
Start
a new ML project
with PSO
Cloud Discover Cloud MVM Cloud Deploy
Machine Learning Lifecycle
62. Confidential & Proprietary
Production Recommendation Solution on GCP
Google Analytics
BigQuery
Google Analytics
360
Customer Web
Application
Web
Server
Application
Server
Database
Server
Rec API
App Engine
Cloud Endpoints
Model Training
Cloud Machine Learning
Orchestration
Cloud Composer
ML Data
Training
Model files
Browser
Client
Mobile /
Tablet Client
63. Confidential & Proprietary
Kurier.at
3rd Largest news provider in Austria
Parent company: 500 web site properties
Google Analytics 360 customer
Outbrain user… Not happy
eDialog
GA Partner
67. complex
solve
The ability to detect patterns in satellite images — such as the
difference between snow and clouds —is critical to Airbus Defense
and Space’s users who depend on highly precise, up-to-date and
reliable information.
Produits Utilisés
Google Cloud Dataflow, BigQuery, Cloud Storage et Cloud Datalab
Industry: Aerospace – Region: France
“In our tests, Google Cloud Machine Learning enabled us
to improve the accuracy and speed at which we analyze
the images captured from our satellites. It solved a
problem that has existed for decades" .
Mathias Ortner, Data Analysis and Image Processing Lead
Solutions
One of our customers, Airbus Defense and Space, tested the use of
Google Cloud Machine Learning to automate the process of
detecting and correcting satellite images that contain imperfections
such as the presence of cloud formations.
problems
opportunities
identify
with advancements such as
Machine Learning
Speed
at which we analyze the images
captured
Accuracy
improved thanks to Machine
Learning
68. Gain
Insights
from images by detecting
individual objects and concepts
Intelligent structure for
30 billion
files managed with powerful
capabilities
Speed up
search and discovery with
image-centric workflows
Enabled accuracy for automation
performance
Industry: Technology – Region: North America
Box customers unlock new value from content as they
automatically classify images and greatly accelerate
business processes.
Improved
Extensive
content management for
customers in every industry
Products Used
Google Cloud Vision API
Solution
Google Cloud Platform and machine learning enable Box to help its
customers manage and gain insight from their image files, and
speed up image-centric processes and workflows.
69. 119languages and dialects
supported
Machine learning that
Enables
applications that react to
what people say
Understand
intent of callers in addition to what
they say
Cutting-edge speech recognition
capabilities
Products Used
Google Cloud Speech API
Industry: Technology – Region: North America
“Machine Learning has changed the game.”
Jeff Lawson, CEO, Twilio
Expanded
support in Twilio Understand
with Google Cloud Speech API
Solution
Google Speech API enables Twilio to make it easier for developers
to build applications that react to what people say during phone
calls, taking callers speech and turn it to text. By adding a layer of
machine intelligence over existing support, customers can bypass
navigating menus and using phone keypads.
70. Machine learning is a core,
transformative way by which we’re
rethinking how we’re doing
everything.
- Sundar Pichai
“
”