Module 3 How Google Does ML

Module Learning ● Acquire the organizational know-how to implement
Objectives machine learning

● Leverage Google's experience to avoid common
pitfalls.
© 2017 Google Inc. All rights reserved. Google and the Google logo are trademarks of Google Inc. All other
company and product names may be trademarks of the respective companies with which they are associated.
Agenda The ML surprise
The secret sauce
The ML and business processes
The path to ML
End of phases deep dive
How Google does ML
TALENT and its uses in organizations
Machine Learning on Google Cloud Platform
© 2017 Google Inc. All rights reserved. Google

and the Google logo are trademarks of Google Inc.
All other company and product names may be
trademarks of the respective companies with
which they are associated.
What is ML?
Machine learning (ML) is the

process of a computer writing
a computer program to
TALENT accomplish a task.
The computer figures out the

“best” program to write by only
looking at a set of examples.
Software Engineers
write program rules
Software Engineering
TALENT
Inputs Outputs
Machine Learning figures
out program rules
Machine learning
LOTS of LOTS of
Inputs Outputs
TALENT
Inputs Outputs
The broccoli surprise
The Broccoli Surprise
TALENT
TALENT
The ML Suprise
Defining KPI’s
Collecting data
TALENT Building infrastructure
Optimizing ML algorithm
Integration
ML Effort Allocation
Defining KPI’s Building infrastructure
Collecting data Optimizing ML algorithm
Integration
TALENT
Expectation
Reality
0.25 0.5 0.75 .1
Then why
are we learning
about ML?
TALENT
Most of the impact
comes along the way
TALENT
Talent / Image layouts added per new Template
Course 1: How Google does ML
Module 3: How Google does ML
Lesson Title: The secret sauce
Format: Screencast
<hello
world>
y = mX + b
TALENT
Get your hands dirty by practicing with technical skills
TALENT
Avoid these top 10 ML pitfalls
Defining KPI’s Collecting data Integration Infrastructure Optimizing ML
1. ML requires just as much software infrastructure

2. No data collected yet
3. Assume the data is ready for use
4. Keep humans in the loop
TALENT 5. Product launch focused on the ML algorithm

6. ML optimizing for the wrong thing
7. Is your ML improving things in the real world
8. Using a pre-trained ML algorithm vs building your own
9. ML algorithms are trained more than once
10. Trying to design your own perception or NLP algorithm
Animate to reveal each of the 10 points as he

company and product names may be trademarks of the respective companies with which they are associated. covers them
Ugh, so that’s the bad news, what's the good news?
Most ML value ML improves If ML is hard, it’s ML is a great

comes along almost everything hard for your differentiator
the way it touches competitors too
Value comes along the way
ML Model
Value Delivered
Future
TALENT
ML Model 3
ML Model 2
ML Model 1
Time
Lesson Title: The ML and Business Processes
Format: Screencast
TALENT No ML
Evolution of a Business Process
Business
Activities
TALENT Organization Customers
Example Call Center Feedback Loop
HR &
CUSTOMER
TRAINING
SURVEYS
REPS CANNED ANSWERS,
PROMOTIONS,
TALENT DECISION TREES
EMPLOYEE
RETRAINING
CALL
CUSTOMER CENTER ANSWERED
QUESTIONS QUESTIONS
General Feedback Loop
INSIGHT
TUNING
GENERATION
OPERATIONAL
PARAMETERS
TALENT UPDATED
INSTRUCTIONS
PROCESS
INPUTS OUTPUTS
Path to ML: The 5 phases

How change happens in phases:
INSIGHT
TUNING
GENERATION
Step 1 - Individual contributor
OPERATIONAL
PARAMETERS
Step 2 - Delegation
UPDATED
Step 3 - Digitization INSTRUCTIONS
Step 4 - Big Data and Analytics
Step 5 - Machine learning PROCESS

INPUTS OUTPUTS

How change happens in phases:
INSIGHT
TUNING
GENERATION
OPERATIONAL
PARAMETERS
Step 2 - Delegation
UPDATED

INPUTS OUTPUTS
ata
How change happens in phases: ML Big D
INSIGHT
TUNING
GENERATION
OPERATIONAL
PARAMETERS
Step 2 - Delegation
UPDATED

INPUTS OUTPUTS
Individual contributor
Delegation
TALENT Digitization
Big Data and Analytics
Machine learning
Google Reception in San Francisco CA
TALENT
Step 2 - Delegation
Store Checker
TALENT
Step 3 - Digitization
ATM
TALENT
ATM
Efficient Car Manufacturing
TALENT
Store Checker
Step 5 - Machine learning
YouTube Recommendation Engine
TALENT
Store Checker
Path to ML: Your turn
INSIGHT
TUNING
GENERATION
OPERATIONAL
PARAMETERS
TALENT UPDATED
INSTRUCTIONS
PROCESS
INPUTS OUTPUTS
Lesson Title: The Path to ML
Format: Screencast
The Path to ML
Step 2 - Delegation
TALENT Step 3 - Digitization
X Step 4 - Big Data and Analytics
Step 5 - Machine learning
Animation note: Fade in arrows as he talks about

going through steps and fade in check and X when he
mentions the word “obstacles”. 0:10 and 0:35
respectively
1 - Individual contributor
2 - Delegation
3 - Digitization
4 - Big Data and Analytics
5 - Machine learning
TALENT
Prototype and try out ideas
2 - Delegation
3 - Digitization
TALENT ● Dangers of skipping this step:

○ Inability to scale
○ Product heads make big, incorrect
assumptions that are hard to change later
● Dangers of lingering too long here:

○ One person gets skilled and then leaves
○ Fail to scale up the process to meet demand
in time
2 - Delegation
3 - Digitization
TALENT
Gently ramp up to include more people

2 - Delegation
3 - Digitization

○ Not forced to formalize the process
○ Inherent diversity in human responses become a
testbed--great product learning opportunity
○ Great ML systems will need humans in the loop
● Dangers of lingering too long here:

○ Paying a high marginal cost to serve each user
○ More voices will say automation isn’t possible
○ Organizational lock-in
2 - Delegation
3 - Digitization
TALENT
Automate mundane parts of the process

2 - Delegation
3 - Digitization

○ You will always need infrastructure
○ IT project and ML success tied and the
whole project will fail if either does
● Dangers of staying here too long:

○ Your competitors are collecting data and
tuning their offers from these new insights
2 - Delegation
3 - Digitization
TALENT
Measure and achieve data-driven success

2 - Delegation
3 - Digitization
TALENT
● Dangers of skipping this step:
○ Unclean data means no ML training
○ You can’t measure success
● Dangers of staying here too long:

○ Limit the complexity of problems you
can solve
2 - Delegation
3 - Digitization
TALENT
Automated
feedback loop
that can outpace
human scale
Lesson Title: End of phases deep dive
Format: Screencast
Reviewing the Path to ML: 5 phases
INDIVIDUAL
CONTRIBUTOR DELEGATION DIGITIZATION BIG DATA MACHINE LEARNING
WHO EXECUTES
THE PROCESS?
INDIVIDUAL
WHO EXECUTES
THE PROCESS?
HOW WERE THE

OPERATIONAL
PARAMETERS
CHOSEN?
INDIVIDUAL
WHO EXECUTES
THE PROCESS?
HOW WERE THE

OPERATIONAL
PARAMETERS
CHOSEN?
HOW WERE THE

PARAMETERS,
FEEDBACK TO
EXECUTION?
TALENT Final Reminders
Don’t Leap into a Fully ML Solution
TALENT
ML is a Journey
ML Model
Value Delivered
Future
TALENT
ML Model 3
ML Model 2
ML Model 1
Time
The Path to ML ...
INSIGHT
TUNING
OPERATIONAL GENERATION
PARAMETERS
UPDATED
TALENT INSTRUCTIONS
PROCESS
INPUT OUTPUTS
...Means automate the blue boxes!
Google can help
INDIVIDUAL
DELEGATION DIGITIZATION BIG DATA ML
CONTRIBUTOR
TALENT Google Cloud Platform
cloud.google.com/training/
TALENT 100% width
(Josh’s content is Next Steps)
cloud.google.com
Images by Connie Zhou

Module 3 How Google Does ML

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Module 3 How Google Does ML

Uploaded by

Copyright:

Available Formats

Module Learning ● Acquire the organizational know-how to implement

Objectives machine learning

The secret sauce

The ML and business processes

End of phases deep dive

Machine Learning on Google Cloud Platform

© 2017 Google Inc. All rights reserved. Google

Machine learning (ML) is the

The computer figures out the

TALENT Building infrastructure

Defining KPI’s Building infrastructure

Collecting data Optimizing ML algorithm

0.25 0.5 0.75 .1

Course 1: How Google does ML

Module 3: How Google does ML

Lesson Title: The secret sauce

1. ML requires just as much software infrastructure

TALENT 5. Product launch focused on the ML algorithm

Animate to reveal each of the 10 points as he

Most ML value ML improves If ML is hard, it’s ML is a great

Course 1: How Google does ML

Module 3: How Google does ML

Lesson Title: The ML and Business Processes

General Feedback Loop

Step 4 - Big Data and Analytics

Step 5 - Machine learning PROCESS

General Feedback Loop

Step 4 - Big Data and Analytics

Step 5 - Machine learning PROCESS

Step 4 - Big Data and Analytics

Step 5 - Machine learning PROCESS

Big Data and Analytics

Google Reception in San Francisco CA

Efficient Car Manufacturing

YouTube Recommendation Engine

Module 3: How Google does ML

Lesson Title: The Path to ML

Step 1 - Individual contributor

X Step 4 - Big Data and Analytics

Step 5 - Machine learning

Animation note: Fade in arrows as he talks about

4 - Big Data and Analytics

Prototype and try out ideas

4 - Big Data and Analytics

TALENT ● Dangers of skipping this step:

● Dangers of lingering too long here:

4 - Big Data and Analytics

Gently ramp up to include more people

4 - Big Data and Analytics

TALENT ● Dangers of skipping this step:

● Dangers of lingering too long here:

4 - Big Data and Analytics

Automate mundane parts of the process

4 - Big Data and Analytics

TALENT ● Dangers of skipping this step:

● Dangers of staying here too long:

4 - Big Data and Analytics

Measure and achieve data-driven success

4 - Big Data and Analytics

● Dangers of staying here too long:

4 - Big Data and Analytics

Course 1: How Google does ML

Module 3: How Google does ML