Skip to main content


6G era is coming to address the urgent growing demand for Zettabyte or even more information exchange among billion of things, connected vehicles, drones, robots, and humans. Current evidence foreshadows the emergence of commercial 6G in 2030 or even a bit earlier. Therefore, on one hand, there is not ample time to pave the way for using 6G and remove left stumbling blocks. On the other hand, secure European sovereignty is intertwined with the development of endemic 6G infrastructures and solutions as a crucial technology. As a glaring example, the COVID-19 pandemic has reminded us of the importance of Information and Communication Technology (ICT) infrastructures, perhaps more than we used to. We all should learn our lesson from this global disaster. 
We will come across a self-contained Artificial Intelligence (AI) ecosystem for 6G. This ecosystem aims to move from human-centric to machine-centric as much as possible in line with zero-touch management principles. Even now, 5G can narrowly manage the current network scale, let alone the expected massive ones in 6G. Thereby, only by automatic solutions can we reach an adaptive optimized real-time network (re)configurations: the more automaticity, the more users' Quality-of-Experience (QoE or QoX) and the more operators' revenue. Wireless AI as a cognitive network technique has emerged by providing a set of mechanisms to configure wireless networks' settings automatically. Take the massive network slicing as one of the most challenging expectations in 6G networks. Roughly speaking, supporting diversified services considering the wide range of requirements and characteristics leads to the construction of multiple logically isolated virtual networks on top of the physical network, called network slicing. Providing intelligent caching strategies notable for mobile edge devices is another example. ML techniques can help implement both proactive and reactive caching strategies. Note that edge resources may be insufficient to stand the wide variety of unavoidable future network services. Therefore, intelligent resource management plays a critical role in future networks. There are many such examples, and the future of 6G relies on wireless AI techniques to overcome these challenges.
Zero-touch 6G network necessitates arming network management approaches with Machine Learning (ML) techniques. Networks face massive data that feed controllers to configure available settings. The data is neither centralized nor homogeneous. For example, a base station can measure network throughput and transmission power, while an end-user can report its speed, own position, and QoE parameters. Network management cannot be careless with a part of this information, and they need all of them to predict accurately and make the right decision. Thereby, using ML in this context necessitates some adaptations. The distributed nature of wireless networks urges ML to head to distributed learning, respecting data privacy. So, Federated Learning (FL) and Split Learning (SL) are the two most practical approaches in this domain. Although there has been a remarkable effort to provide efficient FL and SL techniques, several adaptations and improvements are needed to apply them in the zero-touch 6G network context. This fact needs close cooperation and interaction between researchers in communication and ML communities. 
This proposal aims to provide a trustworthy and green AutoML framework for zero-touch 6G networks. So, integrity and energy efficiency are two fundamental bases for the proposed framework. To the best of our knowledge, most FL techniques assume that all clients are trustworthy, while it is likely to come across untrust clients in such a vast and crowded system like the 6G network. Intentional or unintentional erroneous updates affect not only learning performance but also the effective use of resources. Energy efficiency is not an unknown challenge; however, sometimes, it has been left in shadow. Most investigations have focused on how they can propose high accurate ML techniques, unaware of the importance of energy efficiency. Perhaps the lack of suitable and specialized formulations to evaluate energy consumption for ML techniques has led to the lack of deep interest in this domain. This proposal address this issue as a considerable objective. Furthermore, heterogeneity in data and resources and incentive design for FL techniques in zero-touch 6G implementation is the other objective of this proposal.