The first thing that stands out about the Network Intelligence (NI) for aDAptive and sElf-learning MObile Networks project is its DAEMON acronym.
DAEMON’s in the Details
One look at the project’s logo on its 5G Infrastructure Public Private Partnership (5G-PPP) page should tell visitors all they need to know about what inspired its name, as the “O” actually sports horns on either side. However, like its apparent namesake, there’s more to the project than might meet the eye.
Talking to 6GWorldTM, Project Coordinator Marco Fiore revealed the two sides of DAEMON’s overarching goal, to achieve vector integration of AI with the mobile network of the future and then automate it. With that in mind, the definition of “daemon” in computing, as a program that runs in the background without human intervention, obviously fits as well.
“We feel that current standardisation networks in terms of mobile architecture are not fully aligned with the need to integrate artificial intelligence. On the one hand, we want to complement the current proposals for architectures for beyond 5G in order to fill gaps that prevent artificial intelligence from working optimally,” he said.
“The other side is that artificial intelligence models just taken from the machine-learning community and fed into the network will not work. We need to tune them. So there are two directions, updating the network architecture to accommodate intelligence, but also customising intelligence so it fits networking needs,” he continued.
From 5G to 6G
Networking needs are also ever-changing, especially with the advent of 5G. Talking at a recent Total Telecom webinar Mounir Ladki, CTO of software company MYCOM OSI, broke down the evolution in a B2B context up to this point.
“[Five] years ago, people wanted from us to give them reports: ‘Give me data and I will have people look into them and decide what to do with them,’” he explained, noting that by roughly three years ago the demand had changed: “‘I don’t want you to only give me visibility into the data. I want you to give me outcome-based use cases where you tactically automate some of my processes so I can realize those efficiencies you are talking about.’”
Ladki continued, saying automated networks present a big monetisation opportunity for telecom companies in the next few years. He referenced “advanced connectivity” services, which would include offerings such as Connectivity as a Service, Network as a Service, and network slicing.
“Selling advanced connectivity for B2B, i.e. connectivity services that are real-time, means that you can provision, and you can deliver on demand with [service-level agreements] that you can guarantee in terms of latency, bandwidth, and so on; and you can deliver sometimes at the edge to enable things such as remote factories,” he said. “This is today driving automation demand, i.e., revenue opportunities.”
DAEMON has greater ambitions than opportunities today, though. The Horizon 2020-funded project has an eye on beyond-5G systems, having brought together 12 partners across Europe. IMDEA Networks, at which Fiore is also a Research Associate Professor, is the coordinating body. Other high-profile organisations like NEC Laboratories and Telefonica are also participating in the effort to design algorithms for different NI-assisted functionalities for greater overall performance. The project’s success largely depends on a set of nine key performance indicators. Fiore identified energy efficiency as an especially critical one.
“Networks have begun to consume more energy and this is now a substantial part of the operational expenditure of operators in addition to becoming an issue in terms of global environmental sustainability of the system,” he said. “ I think mobile networks nowadays produce an amount of CO2 that is not far from that of the aviation ecosystem. So, we want to reduce that and, again, network intelligence – that is artificial intelligence applied to networks – is probably the way to go there, because the network is becoming also very complex to manage.
“In order to ensure that our solution meets the energy sustainability targets that are set [externally] we need to move away from the current approach of the human-in-the-loop decision process that is too slow to adapt to the traffic and let the network self-learn and self-manage, also keeping that energy-consumption goal in mind.”
AI in a Networking Context
Fiore said determining how much energy different network functions consume and then optimising them is key. It’s tricky, as functions are run in different settings all the time, using different equipment. Fiore pointed to troubleshooting as a potential area of concern, with interpretability being a limitation of AI in a networking environment that the researchers are looking to address.
“Usually these models are complex, sort of black boxes. You train them. You let the model learn the millions of parameters inside it. You do a validation, you test it, and you see that the model works, but you don’t know how it operates internally,” he said. “In networking, [interpretability] is especially important, because you still want to be able to understand how the model works in order to avoid moving from troubleshooting the network, which is done today by expert human intervention, to letting the AI do the troubleshooting of the network, but then [needing] experts to troubleshoot the AI itself. It’s just moving the problem.”
Fiore called “AI” a very broad term that covers a lot of ground. In fact, in a separate interview, Alex Leadbeater, Chair of the Securing AI Industry Specification Group at the European Telecommunications Standards Institute (ETSI), told 6GWorld part of the challenge of securing AI is how it’s constantly evolving.
“If we think back to the 90s, early 2000s, a degree of what you might term ‘AI-esque’ behaviour in things like automated telephone systems and some of the very early sort of automated bookings lines for cinema tickets started to drift into public perception, but the exact definition of AI, in terms of what is an AI, what different models for AI exist and how are they deployed is essentially very fluid,” Leadbeater had said.
In the case of DAEMON, Fiore said they use the term AI to refer to deep neural networks. However, asked about the obstacles the ever-evolving field of AI presents for DAEMON, Fiore said it’s something the researchers consider regularly and on which they keep up to date. He said most of IMDEA’s partners in the project may have backgrounds in computer networks, but they also interact a great deal with the machine-learning and AI research communities.
“In some cases researchers that are participating come from joint labs like in the case of [Interuniversity Microelectronics Centre; IMEC] in Belgium. They really have a lab that brings together AI experts and machine-learning experts,” he said. “I think one of the things that we’ve been doing in the past, even before DAEMON started, was being very attentive about new paradigms emerging in AI and understanding, experimenting whether they fit needs in the networking field… It’s definitely something we have on our radar.”
Looking Toward (66)6G
Officially having started on the first day of the decade, the project is slated to run until the end of 2023. Considering 6G is only projected to start rolling out near 2030, it could be construed as a significant gap, especially considering the impact DAEMON hopes to have on future 6G systems according to the official press release. Asked about the discrepancy, Fiore offered clarification.
“Many of the things that are foreseen for 5G will not happen at the very beginning of 5G. I think DAEMON is targeting beyond 5G. This will not necessarily be 6G. This may be an evolution of current 5G deployments,” he said. “Many of the things that are in the project are ideas, are concepts. We will need to work on those, develop them and, if we see that they work as expected, we will gain so much new experience that will allow us to define future research directions.
“We are actually participating in working groups organised by 5G-PPP as well as in standardisation bodies like ETSI and many more. The idea is that we want to contribute to the definition of the next generation of mobile networks. DAEMON on its own won’t define 6G… but we want to give a hand towards that. So, I think we are very well-placed. We have many partners who have already contributed to previous standards. So, hopefully, we will be able to have a similar impact with this project.”
Feature image courtesy of Gerd Altmann (via Pixabay).