Go back to programme
Machine learning in communication networks
* Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay
This session starts with an imaginary (?) scenario where generative AI has taken over the networks. Use cases are generated by fine-tuned models, architectures and code, including test code generated by co-pilots and ChatGPT(-likes); digital twins including scenarios and objects are generated by AI-enabled gaming engines; and MLOps pipelines are triggered autonomously. Experts will present their views on the topic, including important challenges such as data collection and modeling in networks, AI-empowered features in 6G and the current state of MLOps. Discussions will focus on questions such as whether networks ready are for this, and identify some of the major areas which need further study including current standards and gaps.
Questions to answer:
1. Given the recent advances in AI/ML, are networks ready for taking advantage of these?
2. How to bring the best advances to networks so that operators and end-users can benefit? what are the integration strategies?
Expected conclusions:
1. More study is needed in the areas of ........
2. Practical challenges in ......
3. Standards gaps in this area are ....
In recent years, we have witnessed several advances in network architectures. Networks are becoming more open and decentralized. On one hand, data is increasingly generated, augmented and processed at the edge of the network. On the other hand, cloud based, scalable compute can enable AI/ML models to infer at high speeds. Sandbox environments would allow for training machine learning models on edge data while keeping it private and secure while controllers and digital twins are spun up on demand. Many algorithms and AI capabilities are now offered as cloud-based services that can be integrated on demand.
Juxtaposing AI-enabled feature advances in future networks such as beam selection, semantic awareness, RIS-enabled radio optimizations, with the architecture evolution, the standards organizations face the difficult task of choosing the right network architectures, placement of functionalities and level of exposure of data while providing application development capabilities to boost opportunities for innovation.
These considerations present new challenges for future networks such as 6G. To benefit from open networks, edge data, and on-demand models, standards organizations need to carefully consider what are the collaborative working models, especially for proof of concepts which can answer some of these questions. Discussions in this session would focus on some of the important considerations in AI-native design and network architectures which can host those designs.
Questions to answer:
1. what are the evolutions in network architectures needed for enabling the integration of AI/ML?
2. What are the techniques to enable generation, collection, utilization of data in networks beyond 5G?
Expected conclusions:
1. Collaborative analysis is needed in the areas of ........with ......
2. PoC studies can mitigate risks in ....
3. The relevant open source studies are .....
4. use cases such as ... has data from ... part of the network, which may be extracted via ... interface, models may be trained in ... and deployed in ...
The 6G vision is shaping up with emerging use cases from metaverse and other user-centric apps. Standards Developing Organizations are marching on with requirements and feature analysis for future networks. This spurs the need for network controllers and agents to manage bandwidth and latency demands from billions of devices. Network platforms spanning software, hardware, cloud and edge with/without GPUs, TPUs, FPGAs and sensors are emerging. Network operators now face a dilemma: what are the right platforms for 6G? Where does enterprise networks come into the picture? Do the major value-adding algorithms, including the network service optimization, become AI-ed and autonomous and move to cloud? The key questions to ponder over are: What platform architectures do we need for this network evolution? What role does open source play in these architectures?
Questions to answer:
1. what are the platform architectures needed for this?
2. whats the role of open source in such architectures?
AI/ML is becoming more and more pervasive in 5G and future networks but the realization options for AI/ML integrated use cases appear fragmented from a method, platforms and design view point. This is partly due to heterogeneity in use case requirements, datasets and models. In addition, we face challenges in bringing innovations from lab to networks due to different perspectives between innovators vs. incumbents, open vs. black boxes, models vs. people (developers, service providers, maintainers, operational engineers), architecture evolution vs. leap-frogging across generations, and cloud vs. edge. This brings to focus the unenviable position of Standards Developing Organizations such as ITU which is required to come up with roadmaps, work items for integration of AI/ML in future networks including 6G.
Questions to answer:
1. what should ITU and other SDOs do?
2. Are there pre-standard, activities to look at?
3. Is the Challenge activities serving the audiences well? what can be done better there?
Expected conclusions:
1. In collaboration with other SDOs, ITU should study ....
2. Challenge should focus on ...type of use cases, expand to ....type of data and models.
3. Papers and code should be published ....more/less... frequently
Speaker(s)
Fathi AbdeldayemLead Technology Standardization and Evolving Technologiesdu UAE
Francesc WilhelmiPostdoctoral Researcher, Radio Systems Research GroupNokia Bell Labs
Karim RabiePrincipal ArchitectRed Hat
James AgajoProfessorFederal University of Technology Minna Nigeria
Fidelis Ikechukwu OnahSenior MemberIEEE
