mit smr aws-2019-slidedeck-combined-rev2 · gold mine of data 18. #techatliberty most large...

44

Upload: dokhue

Post on 04-Jul-2019

213 views

Category:

Documents


0 download

TRANSCRIPT

Gillian ArmstrongTechnologist with Liberty Mutual Insurance

Eric KesslerData scientist and practice manager for AI/ML at Amazon Web Services

John AshleyDirector, global financial services strategy, NVIDIA

Abbie LundbergBusiness technology analyst, Lundberg Media

Implementing AI: From Starting Out to Scaling Up

Gillian Armstrong @virtualgill

TechnologistEmerging Technology

Liberty Mutual

AI in the EnterpriseLearnings from Liberty Mutual’s Journey: A Practitioner’s PerspectiveGillian ArmstrongLiberty Mutual

Back to School?

Start Now

#TechAtLiberty

Rolling out AI in an existing Enterprise

is Slow and Steady work

You can follow, but not fast

8

#TechAtLiberty

You’ll need a Strategy

9

� Education and Awareness� Assess your Existing Systems

− How capable are they of using or integrating with AI?− How comfortable is your company with the use of the Cloud?− Where and in what format is your Data?

� Assess your current in-house Skills− Where will you upskill your existing staff? When and how will you do that?− What will you need to hire for? At what point should you decide that?

� Assess Opportunities and Risks− Where will you start? How will you measure success?

#TechAtLiberty

A Strategy is “a plan of action designed to achieve a major or overall aim.”

Your assessments must lead to a plan of how and when you will prepare your

existing eco-system for the introduction of AI alongside “business as usual.”

This takes time.

10

#TechAtLiberty 11

Build or Buy?

#TechAtLiberty 12

Build AND Buy

#TechAtLiberty 13

Build AND Buy

Pre-packaged solutionsMove Quick and Learn

Pay-as-you-go Cloud Services

#TechAtLiberty 14

Build AND Buy

Pre-packaged solutionsMove Quick and Learn

DifferentiateTotally custom solutions

Pay-as-you-go Cloud Services

Find Safe to Fail Spaces

#TechAtLiberty 16

Internal

Experimental

Customer-facing

Data, Data, Data

#TechAtLiberty

Most large Enterprise companies are sitting on a

gold mine of data

18

#TechAtLiberty

Most large Enterprise companies are sitting on a

gold mine of data

But that’s the problem – it’s still in the mine!

19

#TechAtLiberty

Some Questions to ask…

20

� Where is your data and who can access it?� Do you have the right data for your problems?� Is your data in a format that can be used?

− Will it need to be pre-processed?− What is the data quality?− Does it cover all groups and/or scenarios? Could it have bias?− Are there any legal restrictions?− Are there any ethical concerns?

New Skills

#TechAtLiberty

Some areas to consider…

22

� Data Science� Machine Learning Development� Cloud Development� Business� User Experience� Legal / Contracts / Privacy / Security� PR / Marketing

Continuous Improvement

#TechAtLiberty

Learn Fast, Change Fast24

LEARN / ADJUST

IMPLEMENTMEASURE

John Ashley

Eric KesslerData scientist and practice manager for AI/ML at Amazon Web Services

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

What’s your company’s unfair advantage in ML?

Algorithms Talent Data

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

What’s your company’s unfair advantage in ML?

Algorithms Talent Data

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

ML presents organizational challenges

Inherently iterative and experimental

Dynamic business specifications

Non-standard testing

Feedback loops

Non-standard deployment patterns

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

ML using traditional IT org structures

Business unit Enterprise IT

Data Scientist

• Translates business problem to ML

• Explores & analyses data

• Builds model

Productionize• Captures requirements

and specifications• Develop pipeline and

deploys model• Operates and maintains

model• Manages change requests

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

A vertically integrated team for ML

Model

Pipeline

Product

Data Scientist

ML Engineer

Product Manager

ML

Flex team

UX/frontend engineer

Business SME

Cloud SMEs (Security, Networking, DevOps, etc.)

Core team

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

Accelerate adoption with an ML Center of Excellence

ML CoE

Business unit

Business unit

Provide full stack ML development to BUs across all functions (ML, Dev, Product)

Identify and develop ML roadmaps

Fight attrition through full pipeline of diverse ML projects

Initial adoption of ML

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

Accelerate adoption with an ML Center of Excellence

ML CoE

Business unit

Business unit

Supports decentralization through hiring and ML strategy

Build ML expertise in the BUs

Provide best practice guidance and support

Scaling ML inside of business units

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

Tenets of machine learning in production

Design for scalability, elasticity and stability

Treat data with same rigor as code

Enable full transparency and auditability

Automate as much as possible

Build for consistency by design