deployment of ai solutions in the cloud - flexious€¦ · customized docker yes no training jobs...

40
Deployment of AI solutions in the cloud

Upload: others

Post on 13-Jun-2020

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Deployment of AI solutions in the cloud

Page 2: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

I AM…

Page 3: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

AI IN PRACTICEBuilding blocks

Ready to use APIs offering AI-powered services. Examples include: Speech-To-Text, Object detection, sentiment analysis, etc.

Pre-trained AI models made available through open-source licenses. AI models have both generic and specialized functions.

Frameworks are the base layer on which AI-applications are built.Examples include: Tensorflow, Scikit-Learn, PyTorch, etc.

AI

Services

AI Models

AI Frameworks

Page 4: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes
Page 5: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

AI Services

01

Page 6: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

ML SYSTEMS01

Configuration

Feature Extraction

Data Collection Data Verification

Process

Management Tools

Machine

Resource

Management

Analysis Tools

Serving

Infrastructure

Monitoring

ML

Code

Page 7: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

The Cloud

02

Page 8: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

CLOUD SERVICE PROVIDERS02

AzureML Service

AmazonSageMaker

GoogleAI Platform

Page 9: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

AMAZON SAGEMAKER

Prepare machines and

data

Build modeland image

Train andtest model

Manage and register model

Build image and build endpoint configuration

Deploy endpointconfiguration

Prepare Experiment

Deploy

02

Page 10: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

AZURE ML SERVICE

Build modelTrain and

test modelManage and register

modelBuild image Deploy image

Prepare Experiment

Deploy

Prepare machines and

data

02

Page 11: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

COMPARISON

SageMaker ML Service

Data YES, S3 bucket YES, Storage account

Model trainingCustomized Docker YES NO

Training Jobs YES YES

Model deployment

Vendor lock-in Slight NO

Autoscaling YES YES, Kubernetes

Customized Docker YES EXTRA LAYERS

02

Page 12: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

COMPARISONSpeed

Custom DockerfileLayer caching

Premade DockerfileKubernetes

25’ 70’

02

Page 13: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GOOGLE AI PLATFORM

Prepare dataBuild model as package

Train &test model

Manage and register model

Deploy andmonitor model

Prepare Experiment

Deploy

02

Page 14: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Advice

03

Page 15: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

CONCLUSION03

Page 16: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Azure Machine Learning ServiceGoogle AI Platform

CONCLUSION

Amazon SageMaker

Amazon Web Services Microsoft AzureGoogle Cloud Platform

03

Page 17: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

CONCLUSION

Amazon Web Services

Amazon SageMaker

Google Cloud Platform

Google AI Platform

Microsoft Azure

Azure Machine Learning Service

USER FRIENDLY

FAST / OPTIMIZED

FINISHED

LIGHT VENDOR LOCK-IN

03

Page 18: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

CONCLUSION

Google Cloud Platform

Google AI Platform

Microsoft Azure

Azure Machine Learning Service

Amazon Web Services

Amazon SageMaker

LESS USER FRIENDLY

FAST

IN ACTIVE DEVELOPMENT

DIFFERENT FROM THE REST

03

Page 19: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Google Cloud Platform

Google AI Platform

CONCLUSION

Microsoft Azure

Azure Machine Learning Service

Amazon Web Services

Amazon SageMaker

LESS USER FRIENDLY

SLOW / UNOPTIMIZED

IN ACTIVE DEVELOPMENT

NO VENDOR LOCK-IN

03

Page 20: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

CONCLUSION

Amazon Web Services

Amazon SageMaker

03

Page 21: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

SIZING SERVERS LAB

MACHINE LEARNING &

ARTIFICIAL

INTELLIGENCE

01 DATA ARCHITECTURE02 STRESSTESTING03 PROJECTS04

Make use of Cloud Service Providers

Azure | AWS | GCP

Increase AI implementation speed

and accuracy

Optimize the data architecture

tailored to each project

Stress testing and benchmarking of Big

Data and AI pipelines.

12 years

80+ companies

16 projects

Page 22: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

IoT Infrastructure EngineerAI Engineer

Web App Developer Smart tech & AI Creator

MCT

Page 23: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes
Page 24: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Deployment of AI solutions in the cloud

Page 25: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

In close collaboration with researches, students, and software

developers, DAE Studios aims to create solutions for your company’s

challenges using the many opportunities game technology offers

Page 26: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

D A E E C O S Y S T E M

Page 27: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Opportunity: Expertise Digital Arts & Entertainment

>1200 students

Best game dev school worldwide (2017+2018)

Majority of Flemish studios are either started or staffed by DAE alumni

DAE-Research & XR Lab: innovation through game tech

Opportunity: Flanders Game Valley

Hangar K co-creation hub (Barco, Unilin, Howest, KUL, KBC + 30 start ups)

Multidisciplinary collaborations (Industrial product design, Media & communication tech, DAE,..)

Unesco Creative City, Kortrijk Design Regio

Industry 4.0 – Flanders Make

Opportunity: Game technology

Proven potential of extended reality (Virtual, Mixed & Augmented Reality) within most professional domains

Streamlined pipelines and integrations (CAD, BIM, Scan,.. to game engines)

Massive investments by several industries

Page 28: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECHNOLOGY ACCELERATOR PROGRAM helping start up

companies

GAME TECHNOLOGY TRY OUT PARTNER interactive solutions for your company

GAME TECHNOLOGY INNOVATION AGENT network of fulfilment partners

Page 29: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

HOWEST DAE

Internship student X

Intern

ship

me

nto

r

DAE STUDIOS NV

Project manager

Lead developer

CLIENT

Innovation manager

Innovation teamInternship student Y

Internship student Z

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Page 30: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Case Dewulf

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

How can we give potential customers an impression

of our farming machinery without physically having to

operate them?

Page 31: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Case Dewulf

Page 32: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Can we offer our customers an innovative, intuitive &

interactive web-based game that works on our

databases?

Case Nationale Loterij

Page 33: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Typical pitch of a concept Co creation session Pitch board of directors Representation lotery fair

February March April May June

Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18

ExplorationBRAINSTORM

PrototypingCONCEPT 1 CONCEPT 2 CONCEPT 3 CONCEPT 4

CONCEPT ITERATION

CONCEPT PITCH

Main Concept IterationITERATION SPRINT

ITERATION SPRINT 2 + PITCH

ProductionPRODUCTION SPRINT PRODUCTION SPRINT 2 PRODUCTION POLISH

DELIVERY + PRESENTATION

Case Nationale Loterij

Page 34: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Case Nationale Loterij

Page 35: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

How can we teach our (new) personnel about the

impact and purpose of the many light parameters

used in our spatial planning software?

Case Fluvius

Page 36: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Single point of contact for questions regarding the implementation of game technology

Mapping the concise case and check its technical feasibility

Scouting for subsidy calls and other opportunities

Network of fulfilment partners (start ups or priviliged partners)

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

INSPIRATION

•Game projects

•Group projects

•Graduation work

•Research project

•..

REALISATION

Demonstrator

development

VALORIZATION

Start up company

or

Priviliged partners

HOWEST DAE DAE STUDIOS NV FULFILMENT PARTNER

Page 37: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

accelerator

a venture program that

promotes and aids the

growth of selected

new businesses

DAE Studios Partner programsi.e. Start it @ KBC – imec i.start

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Page 38: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Start up: Rhinox Training

vr training for industry 4.0

Page 39: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

GAME TECH TRY OUT PARTNER INNOVATION AGENT ACCELERATOR PROGRAM

Start up: Play it Safe

game based learning platform for safety and prevention

Page 40: Deployment of AI solutions in the cloud - Flexious€¦ · Customized Docker YES NO Training Jobs YES YES Model deployment Vendor lock-in Slight NO Autoscaling YES YES, Kubernetes

Get in Touch

[email protected]

www.daestudios.be

Are you triggered by the possibilities of game technology?

Do you lack the expertise for developing these solutions in house?

Are you interested in a prototype/vertical slice/demonstrator for your case?