|
Work item:
|
P.MLGuide
|
|
Subject/title:
|
Guide for Development of Machine Learning Based Solutions
|
|
Status:
|
[Carried to next study period]
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
-
|
|
Liaison:
|
ITU-T SG13, SG16
|
|
Supporting members:
|
InfoVista SAS, Opticom, Rohde & Schwarz
|
|
Summary:
|
The ML topic's imminence grew significantly in the telecom industry lately and mainly due to the fact that 5G networks must heavily rely on machine learning; from the intelligently adaptive RAN to real time network slicing adaptation to seamless context aware QoE service delivery and expected transformation of human's demands and perception. It is becoming impetuous for network operators to use machine learning to cost efficiently operate, control and manage their networks. Therefore, in order to remain relevant to the evolving telecom industry, SG 12 needs to adapt to and adopt case by case basis ML based approaches. ML approaches are to some extent use case specific, and these regard the conditions and assumptions within which ML techniques are applied, such as real time with continuous adaptive learning/tuning for non-supervised suited applications (e.g. network quality diagnosis, control and management) or off -line learning for supervised suited applications (e.g. QoE prediction). Regardless of the use case, the following aspects related to topics such as, but not limited to, are addressed in this recommendation guide: training/learning databases integrity and validity (data cleansing); training and validation data bases' split process; machine learning features' selection; ML algorithm's accuracy and consequently its suitability for a specific application; ML overfitting/underfitting test.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2019-05-23 09:34:10
|
|
Last update:
2021-11-03 13:01:24
|
