Contribution to Informa's Telco AI World Summit 2020, talking about driving data-powered innovation and automation at pace and scale. Looking Telco automation and AI frameworks working towards Zero Touch (autonomous network operations) with Zero Defects and Zero Complaints. If you are interested in a copy or want to discuss furher don't be shy reach out.
4. BIG NETWORK DATA.
4
Timing
Action & Reaction
High Velocity
(events/sec)
Large Variety
(e.g., 10k+ event cats)
Very High
Volume
Event Process
approx. <1+>
Alarm per sec
approx. <30+>
Events per
millisecond
Daily (mobile) IP User Plane
Data 750+ Tera Byte
approx. 20 Mega Byte
per millisecond
Lots of Unstructured
data (e.g., ASCI, HEX,
BIN, HTML, …)
Illustration from a big telco network in Europe
milli-seconds to minutes
to hours to days in
response requirements.
7. C
Tool
Assistance
Level 1
Partial
Automation
Level 2
Model-based
Automation
Level 3
High
Autonomy
Level 4
Full
Autonomy
Level 5
Machine-assisted Automation
(Humans necessary although with decreasing
involvement)
Autonomous Operations
(Humans are no longer
involved)
Evolving towards Telco-cloud native
including cognitive (AI/ML) solutions (“salt &
pepper”)
Beyond Automation towards
Level 5 Network Operations
We are here today
8. CENTRALIZED
CONTROL?
(THINKINGPAGE)
1 Brain & 1 Heart
A (peripheral) nervous system with
limited autonomous intelligence.
Not designed for a lot of
redundancy.
Anomaly detection un-
sophisticated.
High degree of
Centralized Orchestration
& Control.
9. VSDE-CENTRALIZED
CONTROL?
(THINKINGPAGE)
9 Brains & 3 Hearts.
High degree of autonomy between
parts.
Highly redundant system design.
Anomaly detection un-sophisticated
(but also less critical).
High degree of De-
centralized
Orchestration & Control.
10. THE NETWORK VIEW
Machine Learning Apps
Big Data Process
North
South
Closed loop
optimization
Illustration
11. CHALLENGESWITHINDUSTRIAL ML.
NON-EXHAUSTIVE.
Machine Learning Agent / Model.
Entanglement
Machine Learning
Systems
mix signals together,
entangling them &
makes isolation of
improvements
largely impossible.
Correction Cascades Undeclared Consumers
m1 m’1 m’’1 m’’’’’’1
Once in place, a correction
cascade can create an
improvement deadlock, as
improving the accuracy of any
individual component actually
leads to system-level
detriments.
correction cascade
See also: D, Sculley et al (2015), “Hidden Technical Debts in Machine Learning”.
Official System
with Model M1 &
Output OM1
OM1
Undeclared
system w. M1
dependency
Undeclared consumers are
expensive at best and dangerous at
worst, because they create a
hidden tight coupling of a given
model to other parts of the
stack/system.
12. “WE HAVE LITTLE EXPERIENCEWITH
INTEGRATING ADVANCED
MACHINE LEARNING
IN INDUSTRIAL SCALE
COMPLEX SYSTEMS,
SUCH AS TELCONETWORKS”
Autonomous
Network
Project
ZSM ISG
ENI ISG
PoCs &
Use Cases
(e.g., China)
Grading
Standards
Autonomous
Network
Levels
13. Access
Data Center = Cloud
Experience n = n + 1
AI Controllers
Learning Agents
Environment
Observations:
Customer interactions.
Actions
Reward
e.g., to achieve
desired outcomes
Many experience iterations
per relevant time unit.
Towards
Intelligent
Automation
Services
Customers
Re-enforcement learning (ML/DL)
Closing the loop.
Dynamic machine learning.
Anomaly detection on infrastructure
as well as Learning Agents / RPAAs.
“Closing the Loop”
ZERO
TOUCH
14. Robotic
Proces Automation
No regret if managed … good ROI
Industrial-scaled AI
High complexity…longer term ROI
Chat bot*(s)
No regret … keep simple
Anomaly detection**
Essential & easy ROI
Specialistic (narrow) AI
Simpler tasks…shorter term ROI
Intelligent Automation
(beyond RPA or RPA meets AI)
Higher complexity…uncertain ROI
Check out (*) for advancing your bots https://www.convercx.com/ & (**) imo one of the most inspiring companies on anomaly detection https://www.anodot.com/
15. THANK YOU!
Acknowledgement
Manythanks tomany industry &corporate colleagues whohave contributed with valuable
insights, discussions &comments throughout this work. I wouldin particularlike tothank
Anodot* forinspiring my thinking on the importance ofanomaly detection in the widest
possible sense ofautonomous network operation Last but notleast, thanks tomy wife Eva
Varadifor her patience during this work.
Contact:
Email: kim.larsen@t-mobile.nl
Linkedin:www.linkedin.com/in/kimklarsen
Blogs:www.aistrategyblog.com & www.techneconomyblog.com
Twitter:@KimKLarsen
(*) https://www.anodot.com/
Hinweis der Redaktion
Inspired by the 5 levels of autonomous driving; Level 1: Driver assistance, Level 2: Partial automation, Level 3: Conditional automation, Level 4: High Automation, Level 5: Full Automation.
PNS (peripheral nervous system) includes motor neurons (autonomous nervous system), parasympathetic nervous system and enteric nervous system.
Giant Pacific Octopus have 3 hearts, 9 brains and blue blood (which is architecturally irrelevant for this story).
Good to keep in mind that not all fruits are low hanging!