5G won't just be fast, it'll do the ML-fuelled self-optimisation thing

That's if the ITU can get telcos talking standards, and it looks like it can

There's a race to apply machine learning to networking applications (especially for 5G), and as so often happens in periods of frenzied development, there's an emerging standardisation gap the International Telecommunications Union hopes to fix.

Following the January Machine Learning and 5G workshop in Geneva, the ITU has launched a focus group to try and identify where standardisation is needed.

Contributors including Deutsche Telekom, Huawei, ZTE, KT and Volkswagen told the workshop they expect machine learning to do some of the heavy lifting in designing, operating and optimising 5G networks.

Phone tower

ITU's latest specs show that 5G is not just a wireless network

READ MORE

The focus group is chaired by Slowomir Stanczak of the Fraunhofer Heinrich Hertz Institut for Telecommunications.

In the video below, Stanczak explains “The main goal of the focus group is to identify standardisation gaps … we can also identify research gaps, and another important topic is related to data formats. We have to clarify which data needs to be collected, do we need to protect the data in the network, questions about the quality of the data.”

Networks already generate huge amounts of data telling operators about user location, movements between cells, and call patterns, but as the focus group's terms of reference (linked here) explains, the group hopes to get better information out of that data with machine learning.

One of the problems the group highlights is that operators want to extract “relevant information” from the network without wasting links (for example, from base stations to the core) passing around unnecessary amounts of operational data.

It's more efficient to use compute resources closer to the edge to run the machine learning, send the insights upstream to the network and (quoting again from the terms of reference) “leverage this knowledge for autonomic network control and management as well as service provisioning”.

As affiliated technologies like software-defined networking and network function virtualisation mature, the ITU also envisages machine learning increasingly taking a hand in automating network control and service provisioning.

The group will be looking at whether members want to standardise formats specifying how to “train, adapt, compress and exchange individual ML algorithms, as well as to ensure that multiple ML algorithms correctly interact with each other.”

Those interactions would have to maintain security, the ITU said, and ensure protection of personal information.

As the ITU wrote, “machine learning applications in communications networking are still very much at their nascent stage of development”.

Nokia Bell Labs' Jakob Hoydis told an ITU workshop “predicted QoS” is a “very promising application” of machine learning.

There are three working groups in the focus group:

  • “Use cases, services, and requirements”, which explains itself;
  • “Data formats and machine learning technologies”, which will categorise machine learning algorithms, and define data formats, and mechanisms for privacy and security;
  • The third, “Machine Learning-aware network architecture”, will analyse network management architectures as networks increasingly interact with machine learning systems.

Stanczak said expanding auto industry connectivity and massively connected sensors will create massive overhead and “a lot of uncertainty in the network”.

“If we proceed [without machine learning] … the solution wouldn't be efficient. Machine learning can help increase efficiency and to enable new applications”. ®

Youtube Video




Biting the hand that feeds IT © 1998–2018