the future of iot - gpu technology...
TRANSCRIPT
The Future of IoT and Intelligent Applications
Hugo Latapie
Cisco Chief Technology and Architecture Office
• IoT and networking use cases
• Rapid and flexible ML/DL PoC to production approach
• ML/DL bridge to symbolic AI -> AGI
• Deployment Frameworks
Agenda
Some IoT Use Cases
Transportation ITS World Congress – Oct 2016
- Analytics tuning: defining zones, adjusting and calibrating cameras, training optimal models- Mandatory integration with VSM - Ability to detect and mitigate image artifacts caused by periodic poor LTE conditions- Automatic recovery (monitor and control client + management agent)- Demographics dashboard
30 live video feeds from 3 tram stations,
fully instrumented tram, 4 cameras at two
ITS demo pods
two live analytics dashboards and real-time zone analytics via a secure REST API that powers the map dashboard
INPUT OUTPUT
ITS World Congress – Oct 2016
Cisco Confidential 6© 2013-2014 Cisco and/or its affiliates. All rights reserved.
Tracking and crowd analytics
Events - Australian Open
• Near field and far field counts
Events - Australian Open
Safety and Security
Tailgating Service
Cisco IoX
VSM Connected Analytics
Video Analytics Agents at the
Edge
Video Analytics Manager in
cloudVideo
Access
Control
System
Location
Services
ISE
Cisco
Prime
Infra.
Data Sources Applications
Lenel Dashboard
Cisco Spark
Tropo
Digital Ceiling
2. Architecture - Enterprise Tailgating Data Flow
Data
Insights
Service Management
Actions/Applications
Automation
Video Badge Reader CMX ISE Cisco Prime Infra.
Crowd Analytics Tailgating Analytics Deep Fusion
Containers
on Cisco
IOx…
Webhooks Registry SchedulerApplication and Policy
Management
…
Digital
Ceiling
Lenel
Dashboard
Security Monitoring and
DispatchDynamic Building Resourcing
…
Dynamic Building HVAC
adjustments
2. Architecture - Enterprise Tailgating Service Architecture
Tailgating Detection
Using Video Analytics Cisco can reduce tailgating % by more than 50%
Enterprise security executives on tailgating:
It is one of the most common and innocent security breaches.
>70% believe they are vulnerable to a security breach from tailgating.
>70% believed that it was very likely a security breach could happen at
their own facility as a result of a tailgating incident.
>25% believed if someone were to tailgate into their facility and commit
a violent crime or theft, the cost of the incident to their organization
would be too high to measure.
Deep Learning - Video Analytics
Tailgating Detection @ Cisco1
**Images have been blurred for privacy reasons
Deep Learning - Video Analytics
Cisco Confidential 14© 2013-2014 Cisco and/or its affiliates. All rights reserved.
security
Physical infrastructure
+ sensors/actuators
Data and
Digital Model
Pndadata
Models
DeepFusion
Explorable Reality / Sim
Sensor data
Video
Explorable Reality / Mimic
IoTDM
Webex/TP/Spark
Paris
Barcelona
Manchester
IoT Smart City
Tour de France
W
K
I
D
Control-Room UX Server
Immersive Lab Pop-Up Lab Mobile
IoT Control Room UX
Sim Platform
INFRA / CLOUD HOSTING
Open / Web data
• SLN
• Enhanced Threat Telemetry and Analytics
• Retina / Spatial Predictive Analytics
• Wireless Service Assurance
• Networking – PNDA + Spark, Tetration, Tesseract
• IT/ Collab /TS /UCS
• Enterprise Data Science Office
Networking and other ML/DL use cases
Queue Analytics
Queue Analytics
• Distributed online learning on heterogenous compute platforms (edge to cloud)
• ML/DL framework agnostic HPC microservices architecture
• Minimalistic nvidia-docker deep learning pipelines
• Optimized DL inference
• SqueezeNet style model compression and optimization
• Multi-modal unsupervised learning
Rapid and flexible ML/DL PoC to Production
ML/DL bridge to symbolic AI -> AGI
Reporting, Dashboards, KPIs, Operations Interface
Policy Engine Alerts Engine Analytic Engine
Telemetry Playback
Geo Location /Real
Time location Mapping
Visualization
Video Event
Management
Real Time/Historic
Reporting
Video Integration
Customer Existing Systems (ERP,
CMX, RFID )
Fault Management & NOC Systems
Ticketing & Workflow Systems
BMS / secondary data
sources
Pre-built integrations
MTConnect BACnet
MQTT AMQP
JSONThings
CAM for IoT
Intelligence
Server
CAM for IoT
Intelligence
Dashboard
and Reports
Cisco CAM for IoT Intelligence
Mobility Services Engine
Identity Services Engine
PrimeNetwork
JasperControlCenter
MobilityExchange
IoT Data Connect
Edge /Fog
Data Layer
Digitization ModelLevels
Application(Reporting, Analytics, Control)
Data Abstraction(Aggregation & Access)
Data Accumulation(Storage)
Edge Computing(Data Element Analysis & Transformation)
Connectivity(Communication & Processing Units)
Physical Devices & Controllers(The “Things” in IoT)
Collaboration & Processes(Involving People & Business Processes)
1
2
3
4
5
6
7
Sensors, Devices, Machines,
Intelligent Edge Nodes of all types
Center
Edge
CA
M fo
r IoT
Inte
lligence
• PNDA brings together a number of open source technologies to provide a simple, scalable open big data analytics Platform for Network Data Analytics
Linux Foundation Collaborative Project based on the Apache ecosystem
What is PNDA?
Orchestration
Controllers
Customer
Devices
Applications
QoE Monitoring
Data
Distribution
Data Store
& Processing
Master Data
Store
PNDA
Batch
Processing
Stream
processing
Real Time
Data Store
Deep H
isto
rical Q
uery
Real T
ime Q
uery
Event Data
Log Data
Metric Data
Network Telemetry
Inventory Topology Geography DNSContext:
Live stream
Capacity Analytics
Billing (Mediation)
Business Intelligence
Fault Analysis
PerfAnalysis
Security and Threat
Analysis
Log Search
SPAN Pod
Service Assurance ArchitectureHigh Level Components
Comput
eNetwork Storage
Span Infra Management Node
Ka
fka
with
Se
cu
rity
Event Analysis &
Srvc Impact
Service
Desk
Inventory
SQM
Inventory
Perf Mgmt
Span Infra Central Node @ Cisco
NGENA OSS + BSS
(COMARCH)
NGENA
Operate
Central Node
Cisco Operate
SPAN PoD
NGENA owns Core
Infrastructure, Service (CPE,
VNF Chain) & Service Platform
(NSO, ESC, CSO) & Central
Node
Cisco/CIS own SPAN
Infrastructure related
Data. Expose only
Summary Events to
NGENA
Log
Search
Metr
ic
Ticket
s
Metric
s
Ka
fka
with
Se
cu
rity
Summary Log Events
Tagging
NGENA Central Node @ NGENA
Metr
ic
Service Assurance
Deliverable
NSO
Service Management Node
CPE
NGENA
GW
NGENA
BB
Syslo
g /
SN
MP
MIB CSR WSA
vASAServic
e
Core Infra
Service
Platform
ESC
IP
SLA
Ta
gg
ing
Ta
gg
ing
CSO
Service Tree
Order Mgmt
Summary• ML/DL tools, frameworks, languages continue to rapidly evolve…
remain agnostic where possible
• Nvidia-docker and unix pipes for inter and intra container communication help with this
• For large scale edge to cloud deployment platform that provides scalability, security, provisioning, and more consider Cisco solutions such as CAM/CIIP/CDP/PNDA.
• What’s next? Neural-symbolic cognitive architectures: opencog.org