Deployment of LLMs at the edge of the 6G network
The repository for this project can be found on Github (CCT-project)
This project was written mainly in LaTeX.
Introduction
This project was a research paper written for the course of Cloud Computing Technologies held by Professors Claudio Ardagna and Marco Anisetti at Università degli studi di Milano.
The idea came from me, and originated from a question I saw asked in a paper: "What if 5G was enough? Do we really need 6G?".
The point of the paper was not to answer this question but to go through the main aspects that have been identified for the next mobile communication standards, therefore the paper is substantially meant as a paper-review of the current research fields linked to 6G. I concentrated on the concept of creating a fusion of 6G and AI in the form of LLMs and I identified some problems and some potential solutions to the problem of doing such a deployment.
In the paper I go through the following steps:
-
Introduction.
-
6G overview: using a whitepaper from Nokia, I go through the 6 key aspect of the new 6G communication standard putting more emphasis on the fact that 6G should be a network that is shaped to the user's need and liking and therefore it's going to make heavy use of AI approaches like LLMs.
-
Opportunities: I use this section to briefly expose opportunities that have been identified in the literature and then I move to explaining my idea of the architecture, that builds upon the spectrum distribution showed in the previous section.
-
Technical limitations: I use this section to expose some of the evident problems connected to the need of having LLMs stored in the network in locations that are as close as possible to the end users and I also propose some solutions found in literature in the field of distributed learning (Federated Learning, Split Learning and LoRA).
In the end I just draw some conclusions and propose some future work that could be done in the field.
The pdf for both the paper and the presentation can be found at the links below