SlideShare a Scribd company logo
TU Wien, Vienna Austria
Distributed Systems Group
https://dsg.tuwien.ac.at
Thomas Rausch @thrauat
Waldemar Hummer
Vinod Muthusamy
Alexander Rashed
Schahram Dustdar
Towards a Serverless Platform for Edge AI
IBM Research AI
HotEdge’19, Renton, WA
2
Drone
With Accelerator
Microsoft Build 2018 // Vision Keynote: https://www.youtube.com/watch?v=rd0Rd8w3FZ0
3
Edge AI Accelerators
Google Edge TPU
NVIDIA Jetson
Intel
Neural Compute Stick
Baidu Kunlun
Microsoft
Project BrainWave
Huawei Atlas
4
AI Operationalization
Hummer et al., ModelOps: Cloud-based Lifecycle Management for Reliable and Trusted AI. IC2E’19.
Process Train Validate Servee
Model
Runtime
Monitoring
Data
Perf.
Process Train Validate Serve
Object
Store
Compute
Cluster
Learning
Cluster
Read
Data
Train
Model
Write
Model
Data
Asset
Trained
Model
ModelOps Platform
5
Serverless Model
{} Event (Request)
Trigger
Node
λ
λ
λ
λ
λ
λ
λ
λ
λ
Resource
λ
def handle(req):
s3 = boto3.client('s3')
with open(tmpfile, 'wb') as f:
s3.download_fileobj('bucket', req['obj'], f)
data = numpy.load(f)
m = train_model(data, req['train_params'])
s3.upload_fileobj(serialize(m), 'bucket', 'model'])
# ...
λλ
Function
Scheduler
Cloud Platform
6
Deviceless Model
{} Event (Request)
Trigger
λλ
Function
Scheduler
??
Edge Cloud
Edge
Edge Cloud
Platform
λ
def handle(req):
s3 = boto3.client('s3')
with open(tmpfile, 'wb') as f:
s3.download_fileobj('bucket', req['obj'], f)
data = numpy.load(f)
m = train_model(data, req['train_params'])
s3.upload_fileobj(serialize(m), 'bucket', 'model'])
# ...
7
 Data and Models as first-class citizens
 Model Selectors
 Policies
 Gates
AI Workflow
Programming Model
 Deviceless function scheduling
 Policy enactment
 Context awareness
 Data locality awareness
Execution Platform
A Serverless Platform for Edge AI
λ
λ
8
@consumes.model(selector={
'type': 'image_classifier',
'data_tags': ['machine_x'],
'accuracy': '>=0.88'
})
def inference(model: Model, request):
data = request['input']
# data prep tasks
prediction = model.estimate(data)
@policy.deadline('2s')
@policy.fn(node = 'user_device',
capability = 'gpu')
@policy.data(network=['company_network'],
strict=True)
@consumes.data(
selector={'urn': 'mnist:data'},
holdout=0.2)
@produce.model(
type='classifier',
urn='mnist:model')
def train(data: Data, request) -> Model:
arr = data.to_ndarray()
return Model(train_model(arr))
@gate.bias(attribute = 'age',
predicate = '<0.8')
@gate.drift(metric = 'confidence',
predicate = '<0.2')
λ
9
@consumes.model(selector={'urn': 'model:base'})
@consumes.data(batch = 100, selector=...)
@produces.model(type='regressor', urn='model:user:{usr}')
@policy.fn(node = 'local')
@policy.data(network = 'local', strict=True)
def refine(model: Model, data: Data):
ndarr = data.to_ndarray() # data artifact API
# transfer learning code
return refined_model
@consumes.model(selector={'urn': 'model:base'})
@consumes.data(batch = 100, selector=...)
@produces.model(type='regressor', urn='model:user:{usr}')
@policy.fn(node = 'local')
@policy.data(network = 'local', strict=True)
def refine(model: Model, data: Data):
ndarr = data.to_ndarray() # data artifact API
# transfer learning code
return refined_model
Network (edge, private)
node:{user}
container
Network (cloud)
f(x)
model u
data
data locality node
model b
λ
Function preprocessor
Scheduler
10
Data Locality Tradeoffs
Cluster Middleware Cluster Middleware Cluster Middleware Cluster Middleware
h
Data
proximity
Container
Image
Deploy the container image to the edge?
OR
Send the data to the cloud?
Edge
11
Skippy
 Built on and Kubernetes
 Kubernetes daemon to discover node capabilities
 Custom Python-based Kubernetes scheduler
● Adds inter-node proximity and data locality as constraints
● Non-monolithic architecture
 Coming to GitHub soon™
λ
12
Preprocess Train Inferenceλ λλ
Scheduler + Simulator: https://git.dsg.tuwien.ac.at/serverless-edge-ai/sched-sim
λ
13
Dipl.-Ing. (MSc), BSc
Thomas Rausch
Research Assistant
TU Wien
Institute of Information Systems Engineering
Argentinierstrasse 8-194-02, Vienna, Austria
T: +43 1 58801-184838
E: trausch@dsg.tuwien.ac.at
https://dsg.tuwien.ac.at/staff/trausch
@consumes.model(selector={'urn': 'model:base'})
@consumes.data(batch = 100, selector=...)
@produces.model(type='regressor', urn='model:user:{usr}')
@policy.fn(node = 'local')
@policy.data(network = 'local', strict=True)
def refine(model: Model, data: Data):
ndarr = data.to_ndarray() # data artifact API
# transfer learning code
return refined_model
Network (edge, private)
node:{user}
container
Network (cloud)
f(x)
model u
data
data locality node
model b
λ
Function preprocessor
Scheduler
{} Event (Request)
Trigger
λλ
Function
Scheduler
Edge Cloud
Edge
Cloud
Platform
λ
def handle(req):
s3 = boto3.client('s3')
with open(tmpfile, 'wb') as f:
s3.download_fileobj('bucket', req['obj'], f)
data = numpy.load(f)
m = train_model(data, req['train_params'])
s3.upload_fileobj(serialize(m), 'bucket', 'model'])
# ...
λ
14
Discussion
●
Correct level of abstraction?
●
API/SDK features?
●
Validation criteria?
●
Deviceless model (does it work?)
●
Transparent data management
●
Scheduler architecture
●
Request routing architecture
●
Proximity and bandwidth monitoring
●
Learning optimal placements
●
Model too high-level for scheduler
●
“Bring-your-own-device” will fail
Feedbacki Controversial pointsii
Open issuesiii Failure risksiv

More Related Content

What's hot

AI @ Microsoft, How we do it and how you can too!
AI @ Microsoft, How we do it and how you can too!AI @ Microsoft, How we do it and how you can too!
AI @ Microsoft, How we do it and how you can too!
Microsoft Tech Community
 
NVIDIA Keynote #GTC21
NVIDIA Keynote #GTC21 NVIDIA Keynote #GTC21
NVIDIA Keynote #GTC21
Alison B. Lowndes
 
Tales of AI agents saving the human race!
Tales of AI agents saving the human race!Tales of AI agents saving the human race!
Tales of AI agents saving the human race!
Alison B. Lowndes
 
Shattering AI Performance Records
Shattering AI Performance RecordsShattering AI Performance Records
Shattering AI Performance Records
NVIDIA
 
Talk on using AI to address some of humanities problems
Talk on using AI to address some of humanities problemsTalk on using AI to address some of humanities problems
Talk on using AI to address some of humanities problems
Alison B. Lowndes
 
Accelerating open science and AI with automated, portable, customizable and r...
Accelerating open science and AI with automated, portable, customizable and r...Accelerating open science and AI with automated, portable, customizable and r...
Accelerating open science and AI with automated, portable, customizable and r...
Grigori Fursin
 
Best Practices for On-Demand HPC in Enterprises
Best Practices for On-Demand HPC in EnterprisesBest Practices for On-Demand HPC in Enterprises
Best Practices for On-Demand HPC in Enterprises
geetachauhan
 
Deep learning @ Edge using Intel's Neural Compute Stick
Deep learning @ Edge using Intel's Neural Compute StickDeep learning @ Edge using Intel's Neural Compute Stick
Deep learning @ Edge using Intel's Neural Compute Stick
geetachauhan
 
HPC Top 5 Stories: May 18th, 2018
HPC Top 5 Stories: May 18th, 2018HPC Top 5 Stories: May 18th, 2018
HPC Top 5 Stories: May 18th, 2018
NVIDIA
 
Fuelling the AI Revolution with Gaming
Fuelling the AI Revolution with GamingFuelling the AI Revolution with Gaming
Fuelling the AI Revolution with Gaming
Alison B. Lowndes
 
NIPS - Deep learning @ Edge using Intel's NCS
NIPS - Deep learning @ Edge using Intel's NCSNIPS - Deep learning @ Edge using Intel's NCS
NIPS - Deep learning @ Edge using Intel's NCS
geetachauhan
 
NVIDIA DataArt IT
NVIDIA DataArt ITNVIDIA DataArt IT
NVIDIA DataArt IT
Alison B. Lowndes
 
Innovation Roundtable
Innovation RoundtableInnovation Roundtable
Innovation Roundtable
Alison B. Lowndes
 
Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...
Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...
Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...
Mahmud Hossain
 
Virtualization and Migration in Cloud - Edge Computing models using OpenStack...
Virtualization and Migration in Cloud - Edge Computing models using OpenStack...Virtualization and Migration in Cloud - Edge Computing models using OpenStack...
Virtualization and Migration in Cloud - Edge Computing models using OpenStack...
Sai praveen Seva
 
Talk on commercialising space data
Talk on commercialising space data Talk on commercialising space data
Talk on commercialising space data
Alison B. Lowndes
 
Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020
Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020
Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020
Toshinori Kujiraoka
 
Intelligent internet of things with Google Cloud
Intelligent internet of things with Google CloudIntelligent internet of things with Google Cloud
Intelligent internet of things with Google Cloud
Henrik Hammer Eliassen
 
AI For Enterprise
AI For EnterpriseAI For Enterprise
AI For Enterprise
NVIDIA
 
Transparent Hardware Acceleration for Deep Learning
Transparent Hardware Acceleration for Deep LearningTransparent Hardware Acceleration for Deep Learning
Transparent Hardware Acceleration for Deep Learning
Indrajit Poddar
 

What's hot (20)

AI @ Microsoft, How we do it and how you can too!
AI @ Microsoft, How we do it and how you can too!AI @ Microsoft, How we do it and how you can too!
AI @ Microsoft, How we do it and how you can too!
 
NVIDIA Keynote #GTC21
NVIDIA Keynote #GTC21 NVIDIA Keynote #GTC21
NVIDIA Keynote #GTC21
 
Tales of AI agents saving the human race!
Tales of AI agents saving the human race!Tales of AI agents saving the human race!
Tales of AI agents saving the human race!
 
Shattering AI Performance Records
Shattering AI Performance RecordsShattering AI Performance Records
Shattering AI Performance Records
 
Talk on using AI to address some of humanities problems
Talk on using AI to address some of humanities problemsTalk on using AI to address some of humanities problems
Talk on using AI to address some of humanities problems
 
Accelerating open science and AI with automated, portable, customizable and r...
Accelerating open science and AI with automated, portable, customizable and r...Accelerating open science and AI with automated, portable, customizable and r...
Accelerating open science and AI with automated, portable, customizable and r...
 
Best Practices for On-Demand HPC in Enterprises
Best Practices for On-Demand HPC in EnterprisesBest Practices for On-Demand HPC in Enterprises
Best Practices for On-Demand HPC in Enterprises
 
Deep learning @ Edge using Intel's Neural Compute Stick
Deep learning @ Edge using Intel's Neural Compute StickDeep learning @ Edge using Intel's Neural Compute Stick
Deep learning @ Edge using Intel's Neural Compute Stick
 
HPC Top 5 Stories: May 18th, 2018
HPC Top 5 Stories: May 18th, 2018HPC Top 5 Stories: May 18th, 2018
HPC Top 5 Stories: May 18th, 2018
 
Fuelling the AI Revolution with Gaming
Fuelling the AI Revolution with GamingFuelling the AI Revolution with Gaming
Fuelling the AI Revolution with Gaming
 
NIPS - Deep learning @ Edge using Intel's NCS
NIPS - Deep learning @ Edge using Intel's NCSNIPS - Deep learning @ Edge using Intel's NCS
NIPS - Deep learning @ Edge using Intel's NCS
 
NVIDIA DataArt IT
NVIDIA DataArt ITNVIDIA DataArt IT
NVIDIA DataArt IT
 
Innovation Roundtable
Innovation RoundtableInnovation Roundtable
Innovation Roundtable
 
Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...
Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...
Aura: An IoT based Cloud Infrastructure for Localized Mobile Computation Outs...
 
Virtualization and Migration in Cloud - Edge Computing models using OpenStack...
Virtualization and Migration in Cloud - Edge Computing models using OpenStack...Virtualization and Migration in Cloud - Edge Computing models using OpenStack...
Virtualization and Migration in Cloud - Edge Computing models using OpenStack...
 
Talk on commercialising space data
Talk on commercialising space data Talk on commercialising space data
Talk on commercialising space data
 
Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020
Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020
Arm Neoverse solutions @Graviton2-AWS Japan Webinar Oct2020
 
Intelligent internet of things with Google Cloud
Intelligent internet of things with Google CloudIntelligent internet of things with Google Cloud
Intelligent internet of things with Google Cloud
 
AI For Enterprise
AI For EnterpriseAI For Enterprise
AI For Enterprise
 
Transparent Hardware Acceleration for Deep Learning
Transparent Hardware Acceleration for Deep LearningTransparent Hardware Acceleration for Deep Learning
Transparent Hardware Acceleration for Deep Learning
 

Similar to Towards a Serverless Platform for Edge AI

AWS re:Invent 2016 : announcement, technical demos and feedbacks
AWS re:Invent 2016 : announcement, technical demos and feedbacksAWS re:Invent 2016 : announcement, technical demos and feedbacks
AWS re:Invent 2016 : announcement, technical demos and feedbacks
Emmanuel Quentin
 
DevSum'15 : Microsoft Azure and Things
DevSum'15 : Microsoft Azure and ThingsDevSum'15 : Microsoft Azure and Things
DevSum'15 : Microsoft Azure and Things
Thomas Conté
 
Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...
Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...
Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...
Shift Conference
 
Build and deploy PyTorch models with Azure Machine Learning - Henk - CCDays
Build and deploy PyTorch models with Azure Machine Learning - Henk - CCDaysBuild and deploy PyTorch models with Azure Machine Learning - Henk - CCDays
Build and deploy PyTorch models with Azure Machine Learning - Henk - CCDays
CodeOps Technologies LLP
 
Viktor Tsykunov: Azure Machine Learning Service
Viktor Tsykunov: Azure Machine Learning ServiceViktor Tsykunov: Azure Machine Learning Service
Viktor Tsykunov: Azure Machine Learning Service
Lviv Startup Club
 
Siddhi - cloud-native stream processor
Siddhi - cloud-native stream processorSiddhi - cloud-native stream processor
Siddhi - cloud-native stream processor
Sriskandarajah Suhothayan
 
Ultra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & Alluxio
Ultra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & AlluxioUltra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & Alluxio
Ultra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & Alluxio
Alluxio, Inc.
 
Azure machine learning service
Azure machine learning serviceAzure machine learning service
Azure machine learning service
Ruth Yakubu
 
What Is Happening At The Edge
What Is Happening At The EdgeWhat Is Happening At The Edge
What Is Happening At The Edge
Amazon Web Services
 
Computer project
Computer projectComputer project
Computer project
Pranav Nedungadi
 
Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]
Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]
Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]
Animesh Singh
 
Vert.x devoxx london 2013
Vert.x devoxx london 2013Vert.x devoxx london 2013
Vert.x devoxx london 2013
Stuart (Pid) Williams
 
Serverless machine learning architectures at Helixa
Serverless machine learning architectures at HelixaServerless machine learning architectures at Helixa
Serverless machine learning architectures at Helixa
Data Science Milan
 
End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017
End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017
End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017
StampedeCon
 
Azure Digital Twins.pdf
Azure Digital Twins.pdfAzure Digital Twins.pdf
Azure Digital Twins.pdf
Tomasz Kopacz
 
Data Science und Machine Learning im Kubernetes-Ökosystem
Data Science und Machine Learning im Kubernetes-ÖkosystemData Science und Machine Learning im Kubernetes-Ökosystem
Data Science und Machine Learning im Kubernetes-Ökosystem
inovex GmbH
 
Deploy Deep Learning Models with TensorFlow + Lambda
Deploy Deep Learning Models with TensorFlow + LambdaDeploy Deep Learning Models with TensorFlow + Lambda
Deploy Deep Learning Models with TensorFlow + Lambda
Greg Werner
 
S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...
S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...
S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...
Henry Saputra
 
Getting started with open stack
Getting started with open stackGetting started with open stack
Getting started with open stack
Dan Radez
 
OSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at Netflix
OSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at NetflixOSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at Netflix
OSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at Netflix
Manish Pandit
 

Similar to Towards a Serverless Platform for Edge AI (20)

AWS re:Invent 2016 : announcement, technical demos and feedbacks
AWS re:Invent 2016 : announcement, technical demos and feedbacksAWS re:Invent 2016 : announcement, technical demos and feedbacks
AWS re:Invent 2016 : announcement, technical demos and feedbacks
 
DevSum'15 : Microsoft Azure and Things
DevSum'15 : Microsoft Azure and ThingsDevSum'15 : Microsoft Azure and Things
DevSum'15 : Microsoft Azure and Things
 
Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...
Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...
Shift Remote AI: Build and deploy PyTorch Models with Azure Machine Learning ...
 
Build and deploy PyTorch models with Azure Machine Learning - Henk - CCDays
Build and deploy PyTorch models with Azure Machine Learning - Henk - CCDaysBuild and deploy PyTorch models with Azure Machine Learning - Henk - CCDays
Build and deploy PyTorch models with Azure Machine Learning - Henk - CCDays
 
Viktor Tsykunov: Azure Machine Learning Service
Viktor Tsykunov: Azure Machine Learning ServiceViktor Tsykunov: Azure Machine Learning Service
Viktor Tsykunov: Azure Machine Learning Service
 
Siddhi - cloud-native stream processor
Siddhi - cloud-native stream processorSiddhi - cloud-native stream processor
Siddhi - cloud-native stream processor
 
Ultra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & Alluxio
Ultra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & AlluxioUltra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & Alluxio
Ultra Fast Deep Learning in Hybrid Cloud using Intel Analytics Zoo & Alluxio
 
Azure machine learning service
Azure machine learning serviceAzure machine learning service
Azure machine learning service
 
What Is Happening At The Edge
What Is Happening At The EdgeWhat Is Happening At The Edge
What Is Happening At The Edge
 
Computer project
Computer projectComputer project
Computer project
 
Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]
Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]
Hybrid Cloud, Kubeflow and Tensorflow Extended [TFX]
 
Vert.x devoxx london 2013
Vert.x devoxx london 2013Vert.x devoxx london 2013
Vert.x devoxx london 2013
 
Serverless machine learning architectures at Helixa
Serverless machine learning architectures at HelixaServerless machine learning architectures at Helixa
Serverless machine learning architectures at Helixa
 
End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017
End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017
End-to-end Big Data Projects with Python - StampedeCon Big Data Conference 2017
 
Azure Digital Twins.pdf
Azure Digital Twins.pdfAzure Digital Twins.pdf
Azure Digital Twins.pdf
 
Data Science und Machine Learning im Kubernetes-Ökosystem
Data Science und Machine Learning im Kubernetes-ÖkosystemData Science und Machine Learning im Kubernetes-Ökosystem
Data Science und Machine Learning im Kubernetes-Ökosystem
 
Deploy Deep Learning Models with TensorFlow + Lambda
Deploy Deep Learning Models with TensorFlow + LambdaDeploy Deep Learning Models with TensorFlow + Lambda
Deploy Deep Learning Models with TensorFlow + Lambda
 
S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...
S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...
S8277 - Introducing Krylov: AI Platform that Empowers eBay Data Science and E...
 
Getting started with open stack
Getting started with open stackGetting started with open stack
Getting started with open stack
 
OSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at Netflix
OSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at NetflixOSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at Netflix
OSCON 2014 - API Ecosystem with Scala, Scalatra, and Swagger at Netflix
 

More from Thomas Rausch

Test cloud application deployments locally and in CI without staging environm...
Test cloud application deployments locally and in CI without staging environm...Test cloud application deployments locally and in CI without staging environm...
Test cloud application deployments locally and in CI without staging environm...
Thomas Rausch
 
Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...
Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...
Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...
Thomas Rausch
 
Portable Energy-Aware Cluster-Based Edge Computers
Portable Energy-Aware Cluster-Based Edge ComputersPortable Energy-Aware Cluster-Based Edge Computers
Portable Energy-Aware Cluster-Based Edge Computers
Thomas Rausch
 
EMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing Applications
EMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing ApplicationsEMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing Applications
EMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing Applications
Thomas Rausch
 
Message-Oriented Middleware for Edge Computing Applications
Message-Oriented Middleware for Edge Computing ApplicationsMessage-Oriented Middleware for Edge Computing Applications
Message-Oriented Middleware for Edge Computing Applications
Thomas Rausch
 
An Empirical Analysis of Build Failures in the Continuous Integration Workflo...
An Empirical Analysis of Build Failures in the Continuous Integration Workflo...An Empirical Analysis of Build Failures in the Continuous Integration Workflo...
An Empirical Analysis of Build Failures in the Continuous Integration Workflo...
Thomas Rausch
 
Build Failure Prediction in Continuous Integration Workflows
Build Failure Prediction in Continuous Integration WorkflowsBuild Failure Prediction in Continuous Integration Workflows
Build Failure Prediction in Continuous Integration Workflows
Thomas Rausch
 
Git Introduction Tutorial
Git Introduction TutorialGit Introduction Tutorial
Git Introduction Tutorial
Thomas Rausch
 

More from Thomas Rausch (8)

Test cloud application deployments locally and in CI without staging environm...
Test cloud application deployments locally and in CI without staging environm...Test cloud application deployments locally and in CI without staging environm...
Test cloud application deployments locally and in CI without staging environm...
 
Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...
Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...
Synthesizing Plausible Infrastructure Configurations for Evaluating Edge Comp...
 
Portable Energy-Aware Cluster-Based Edge Computers
Portable Energy-Aware Cluster-Based Edge ComputersPortable Energy-Aware Cluster-Based Edge Computers
Portable Energy-Aware Cluster-Based Edge Computers
 
EMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing Applications
EMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing ApplicationsEMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing Applications
EMMA: Distributed QoS-Aware MQTT Middleware for Edge Computing Applications
 
Message-Oriented Middleware for Edge Computing Applications
Message-Oriented Middleware for Edge Computing ApplicationsMessage-Oriented Middleware for Edge Computing Applications
Message-Oriented Middleware for Edge Computing Applications
 
An Empirical Analysis of Build Failures in the Continuous Integration Workflo...
An Empirical Analysis of Build Failures in the Continuous Integration Workflo...An Empirical Analysis of Build Failures in the Continuous Integration Workflo...
An Empirical Analysis of Build Failures in the Continuous Integration Workflo...
 
Build Failure Prediction in Continuous Integration Workflows
Build Failure Prediction in Continuous Integration WorkflowsBuild Failure Prediction in Continuous Integration Workflows
Build Failure Prediction in Continuous Integration Workflows
 
Git Introduction Tutorial
Git Introduction TutorialGit Introduction Tutorial
Git Introduction Tutorial
 

Recently uploaded

Toxic effects of heavy metals : Lead and Arsenic
Toxic effects of heavy metals : Lead and ArsenicToxic effects of heavy metals : Lead and Arsenic
Toxic effects of heavy metals : Lead and Arsenic
sanjana502982
 
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
University of Maribor
 
20240520 Planning a Circuit Simulator in JavaScript.pptx
20240520 Planning a Circuit Simulator in JavaScript.pptx20240520 Planning a Circuit Simulator in JavaScript.pptx
20240520 Planning a Circuit Simulator in JavaScript.pptx
Sharon Liu
 
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills MN
 
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
David Osipyan
 
DMARDs Pharmacolgy Pharm D 5th Semester.pdf
DMARDs Pharmacolgy Pharm D 5th Semester.pdfDMARDs Pharmacolgy Pharm D 5th Semester.pdf
DMARDs Pharmacolgy Pharm D 5th Semester.pdf
fafyfskhan251kmf
 
Deep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless ReproducibilityDeep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless Reproducibility
University of Rennes, INSA Rennes, Inria/IRISA, CNRS
 
platelets_clotting_biogenesis.clot retractionpptx
platelets_clotting_biogenesis.clot retractionpptxplatelets_clotting_biogenesis.clot retractionpptx
platelets_clotting_biogenesis.clot retractionpptx
muralinath2
 
Phenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvementPhenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvement
IshaGoswami9
 
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Studia Poinsotiana
 
ESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptxESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptx
PRIYANKA PATEL
 
Leaf Initiation, Growth and Differentiation.pdf
Leaf Initiation, Growth and Differentiation.pdfLeaf Initiation, Growth and Differentiation.pdf
Leaf Initiation, Growth and Differentiation.pdf
RenuJangid3
 
Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.
Nistarini College, Purulia (W.B) India
 
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptxThe use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
MAGOTI ERNEST
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
yqqaatn0
 
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
University of Maribor
 
What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.
moosaasad1975
 
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Sérgio Sacani
 
Lateral Ventricles.pdf very easy good diagrams comprehensive
Lateral Ventricles.pdf very easy good diagrams comprehensiveLateral Ventricles.pdf very easy good diagrams comprehensive
Lateral Ventricles.pdf very easy good diagrams comprehensive
silvermistyshot
 
SAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdfSAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdf
KrushnaDarade1
 

Recently uploaded (20)

Toxic effects of heavy metals : Lead and Arsenic
Toxic effects of heavy metals : Lead and ArsenicToxic effects of heavy metals : Lead and Arsenic
Toxic effects of heavy metals : Lead and Arsenic
 
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
 
20240520 Planning a Circuit Simulator in JavaScript.pptx
20240520 Planning a Circuit Simulator in JavaScript.pptx20240520 Planning a Circuit Simulator in JavaScript.pptx
20240520 Planning a Circuit Simulator in JavaScript.pptx
 
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
 
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
 
DMARDs Pharmacolgy Pharm D 5th Semester.pdf
DMARDs Pharmacolgy Pharm D 5th Semester.pdfDMARDs Pharmacolgy Pharm D 5th Semester.pdf
DMARDs Pharmacolgy Pharm D 5th Semester.pdf
 
Deep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless ReproducibilityDeep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless Reproducibility
 
platelets_clotting_biogenesis.clot retractionpptx
platelets_clotting_biogenesis.clot retractionpptxplatelets_clotting_biogenesis.clot retractionpptx
platelets_clotting_biogenesis.clot retractionpptx
 
Phenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvementPhenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvement
 
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
 
ESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptxESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptx
 
Leaf Initiation, Growth and Differentiation.pdf
Leaf Initiation, Growth and Differentiation.pdfLeaf Initiation, Growth and Differentiation.pdf
Leaf Initiation, Growth and Differentiation.pdf
 
Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.
 
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptxThe use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
 
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
 
What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.
 
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
 
Lateral Ventricles.pdf very easy good diagrams comprehensive
Lateral Ventricles.pdf very easy good diagrams comprehensiveLateral Ventricles.pdf very easy good diagrams comprehensive
Lateral Ventricles.pdf very easy good diagrams comprehensive
 
SAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdfSAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdf
 

Towards a Serverless Platform for Edge AI

  • 1. TU Wien, Vienna Austria Distributed Systems Group https://dsg.tuwien.ac.at Thomas Rausch @thrauat Waldemar Hummer Vinod Muthusamy Alexander Rashed Schahram Dustdar Towards a Serverless Platform for Edge AI IBM Research AI HotEdge’19, Renton, WA
  • 2. 2 Drone With Accelerator Microsoft Build 2018 // Vision Keynote: https://www.youtube.com/watch?v=rd0Rd8w3FZ0
  • 3. 3 Edge AI Accelerators Google Edge TPU NVIDIA Jetson Intel Neural Compute Stick Baidu Kunlun Microsoft Project BrainWave Huawei Atlas
  • 4. 4 AI Operationalization Hummer et al., ModelOps: Cloud-based Lifecycle Management for Reliable and Trusted AI. IC2E’19. Process Train Validate Servee Model Runtime Monitoring Data Perf. Process Train Validate Serve Object Store Compute Cluster Learning Cluster Read Data Train Model Write Model Data Asset Trained Model ModelOps Platform
  • 5. 5 Serverless Model {} Event (Request) Trigger Node λ λ λ λ λ λ λ λ λ Resource λ def handle(req): s3 = boto3.client('s3') with open(tmpfile, 'wb') as f: s3.download_fileobj('bucket', req['obj'], f) data = numpy.load(f) m = train_model(data, req['train_params']) s3.upload_fileobj(serialize(m), 'bucket', 'model']) # ... λλ Function Scheduler Cloud Platform
  • 6. 6 Deviceless Model {} Event (Request) Trigger λλ Function Scheduler ?? Edge Cloud Edge Edge Cloud Platform λ def handle(req): s3 = boto3.client('s3') with open(tmpfile, 'wb') as f: s3.download_fileobj('bucket', req['obj'], f) data = numpy.load(f) m = train_model(data, req['train_params']) s3.upload_fileobj(serialize(m), 'bucket', 'model']) # ...
  • 7. 7  Data and Models as first-class citizens  Model Selectors  Policies  Gates AI Workflow Programming Model  Deviceless function scheduling  Policy enactment  Context awareness  Data locality awareness Execution Platform A Serverless Platform for Edge AI λ λ
  • 8. 8 @consumes.model(selector={ 'type': 'image_classifier', 'data_tags': ['machine_x'], 'accuracy': '>=0.88' }) def inference(model: Model, request): data = request['input'] # data prep tasks prediction = model.estimate(data) @policy.deadline('2s') @policy.fn(node = 'user_device', capability = 'gpu') @policy.data(network=['company_network'], strict=True) @consumes.data( selector={'urn': 'mnist:data'}, holdout=0.2) @produce.model( type='classifier', urn='mnist:model') def train(data: Data, request) -> Model: arr = data.to_ndarray() return Model(train_model(arr)) @gate.bias(attribute = 'age', predicate = '<0.8') @gate.drift(metric = 'confidence', predicate = '<0.2') λ
  • 9. 9 @consumes.model(selector={'urn': 'model:base'}) @consumes.data(batch = 100, selector=...) @produces.model(type='regressor', urn='model:user:{usr}') @policy.fn(node = 'local') @policy.data(network = 'local', strict=True) def refine(model: Model, data: Data): ndarr = data.to_ndarray() # data artifact API # transfer learning code return refined_model @consumes.model(selector={'urn': 'model:base'}) @consumes.data(batch = 100, selector=...) @produces.model(type='regressor', urn='model:user:{usr}') @policy.fn(node = 'local') @policy.data(network = 'local', strict=True) def refine(model: Model, data: Data): ndarr = data.to_ndarray() # data artifact API # transfer learning code return refined_model Network (edge, private) node:{user} container Network (cloud) f(x) model u data data locality node model b λ Function preprocessor Scheduler
  • 10. 10 Data Locality Tradeoffs Cluster Middleware Cluster Middleware Cluster Middleware Cluster Middleware h Data proximity Container Image Deploy the container image to the edge? OR Send the data to the cloud? Edge
  • 11. 11 Skippy  Built on and Kubernetes  Kubernetes daemon to discover node capabilities  Custom Python-based Kubernetes scheduler ● Adds inter-node proximity and data locality as constraints ● Non-monolithic architecture  Coming to GitHub soon™ λ
  • 12. 12 Preprocess Train Inferenceλ λλ Scheduler + Simulator: https://git.dsg.tuwien.ac.at/serverless-edge-ai/sched-sim λ
  • 13. 13 Dipl.-Ing. (MSc), BSc Thomas Rausch Research Assistant TU Wien Institute of Information Systems Engineering Argentinierstrasse 8-194-02, Vienna, Austria T: +43 1 58801-184838 E: trausch@dsg.tuwien.ac.at https://dsg.tuwien.ac.at/staff/trausch @consumes.model(selector={'urn': 'model:base'}) @consumes.data(batch = 100, selector=...) @produces.model(type='regressor', urn='model:user:{usr}') @policy.fn(node = 'local') @policy.data(network = 'local', strict=True) def refine(model: Model, data: Data): ndarr = data.to_ndarray() # data artifact API # transfer learning code return refined_model Network (edge, private) node:{user} container Network (cloud) f(x) model u data data locality node model b λ Function preprocessor Scheduler {} Event (Request) Trigger λλ Function Scheduler Edge Cloud Edge Cloud Platform λ def handle(req): s3 = boto3.client('s3') with open(tmpfile, 'wb') as f: s3.download_fileobj('bucket', req['obj'], f) data = numpy.load(f) m = train_model(data, req['train_params']) s3.upload_fileobj(serialize(m), 'bucket', 'model']) # ... λ
  • 14. 14 Discussion ● Correct level of abstraction? ● API/SDK features? ● Validation criteria? ● Deviceless model (does it work?) ● Transparent data management ● Scheduler architecture ● Request routing architecture ● Proximity and bandwidth monitoring ● Learning optimal placements ● Model too high-level for scheduler ● “Bring-your-own-device” will fail Feedbacki Controversial pointsii Open issuesiii Failure risksiv