Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Common Data and Intelligence Layer
Edgar Ramos (Ericsson)Roberto Morabito (Princeton U.)Marie-José Montpetit ( U. Concordia)Eve Schooler ( Intel)
April 2020
Centralized vs. Distributed Computing Architectures
From centralized to
distributed
Data sources
Execution environment
Applications
Centralized Decentralized Distributed
Level of interoperability dependency
— g Inference Execution DistributionAI Service Functional Distribution
Intelligence provisioning (Model distribution)
Agent Functional Distribution
Training Execution Distribution
Computational Intelligence Distribution Aspects
Swarm
HierarchicalShape analyser
Colour analyser
Fruit Recognition
Edge-Computing Fabric Stacks Mapping
Knowledge
Data
Intend &purpose
Sem
antic
s
Communication
Sens
ing
&
Actu
atio
n ca
pabi
litie
s
AgentInteraction
HW/SW Platform
Orc
hest
ratio
n Local Data Sources/Repository
Local Processing
Common Data/Intelligence Layer
Orc
hest
ratio
n
Services & Functions
Peering
Policies Handling
Services Exposure
LCM
Applications
What to do or what does it need from others and how is configured or directed about what to do (or orchestrated)
General and domain specific vocabularies and ontologies that can interoperate
Exchange of processed information according to interaction
(filtering, searching, contextualizing, negotiating, etc.)
Measurement, reports, identifiers, digital representations, etc. Considerations such as data nomadism
(data moving with physical entities)
Multiple topologies and communication modes(D2D, D2Nw, Nw2Nw)
Interaction and negotiation process( cooperate, compete, persuade, block)
Edge architecture stack Functional stack
Monitoring
• Environmental
• Climate
• Irrigation/
nutriment
Local image
processing/sensor
fusion
IoT broker
Local databases
Local decision
Formatting
Match/action
Query/response
Store/process
Data Communications
Data CaptureEdge Computing Cloud Computing
LAN• WSN
• WIFI
• Bluetooth
• Zigbee
• LORA
• RFID
Sensors
• Temperature (
• Humidity
• Moisture
• Ph
• CO2
Cameras
• IR
• Visual
Operator(s)
Farm management
Application/API
management
IoT Broker
Event databases/logs
Cloud-based image
processing/training
Datasets
Remote decision
Status and diagnostics
Cyberphysical security
PLC/Automation
Middleware
Application services
• Phone
• PC/Mac
• Tablet
Local ML/AI
Operator(s)
Alerts/critical
events
Data Interpretation
Ale
rts/
crit
ica
l
eve
nts
CADM
IP Protocols⁃ UDP
• TCP
• HTTPS
• QUIC IoT
• IoT specific
Data Formats & semantics⁃OneDM
•SENML
•LWM2M
•WoT
•JSON-LD
WAN•Cellular
•Fiber +WIFI
•HFC+WIFI
•xDSL+WIFI
Com
mon
Dat
a La
yer i
n an
Ap
plic
atio
n co
ntex
t
Summary
• Interoperability requirements analysis (according to different scenarios)• Common Data Layer taxonomy• "Common Data Layer for COIN" draft
• Possible legacy with the "Edge Data Discovery for COIN" draft
Additional References:Distributing Intelligence to the Edge and BeyondIntelligence Stratum for IoT. Architecture Requirements and Functions
• One of the main challenges in computational intelligence is the lack of interoperability between AI components• Such issue is clearly accentuated at the data layer (e.g., incompatible data
types and formats, heterogeneous APIs and platforms required to execute the AI models, inconsistent life-cycle management and policies)• A Common Intelligence-Data Layer is needed.
Next Steps