My keynote from this weeks (June 12, 2018) 5G World conference in London. I focus here on AI for Telcos and in particular what can be done in the network and IT parts of the network. Making the point that there remains a lot of low hanging fruits which are simpler and easier to solve than present day image and NLP challenges. It is amazing how much you can do with fairly simple ML architectures with out going to state-of-art DL frameworks.
4. More than 70% of consumers do not trust
companies have their best interest in mind!
SurveyMonkey “Millennials – Digital Challenges & Human Answers Survey “ (March 2018).
8. HOW AI? NEED TO
“CLEAN” DATA!
QUALITY
GOALS!
(to train the model, normally not as heavy to run it) (need to beat “Flipping a Coin” or Majority “Vote”)
IT STARTS
HERE!
Train Test
COMPUTING
POWER
LOTS
OF DATA!
(the more data, the higher quality should result & the
less complexity required … in general)
MODEL /
ARCHITECTURE
13. Access
Data Center
Experience n = n + 1
Telco AIs
Learning Agents
EnvironmentObservations:
Data Center, Access
Device, Services
Customers, Etc..
Actions
e.g., driving SON
Reward
e.g., to achieve
desired outcomes
Millions of Experience
iterations per hour.
3Z TELCO
Services
Possible ingredients:
RNN with Reinforcement learning.
LSTM, anomaly detection, ML + FFT, …
POFCNN (plain-old fully-connected nn)
14. TELCO AI
Machine Intelligent
Prevent Tool Basic Research
NG-SON RNN w. re-
enforcement learning
Anodot
There are a lot of
low hanging ML fruits
in Telco
Network & IT.
Illustrations (not exhaustive)
Anomaly Detection
RFMLS & Spectrum
Collaboration Challenge
15. NETWORK OPERATIONS
(sub)milliseconds to minutes
NETWORK OPTIMIZATION
(sub)seconds to minutes
PROACTIVE USER EXPERIENCE
(sub)milliseconds to (sub)seconds
COMPLAINT MANAGEMENT
minutes to hours
NETWORK MAINTENANCE
hours to months
16. THANK YOU!
Acknowledgement
Many thanks many colleagues who have
contributed with valuable insights, discussions
& comments throughout this work.
Also I would like to thank my wife Eva Varadi
for her patience during this work.
Contact:
Email: kim.larsen@telekom.hu
Linkedin: www.linkedin.com/in/kimklarsen
Blogs: www.aistrategyblog.com & www.techneconomyblog.com
Twitter: @KimKLarsen