AI-driven Intent-based Optimization with Algorithmic Machine Learning for 6G Networks

Purpose and goal

This thesis project aims at providing a solid foundation for tackling a critical challenge in 6G networks: optimizing resource allocation in large-scale networks. Current heuristic-based algorithms for network management suffer from sub-optimality (which leads to higher energy consumption), and current learning-based approaches are problem-specific and solution-tailored. We aim at exploring machine learning methods that enable intent-driven optimization for resource allocation problems.

Expected effects and result

We will explore the combination of Graph Neural Networks (GNNs) and Large Language Models (LLMs) for intent-based optimization. Since many resource allocation problems can be traced back to traditional combinatorial optimization problems (COPs), we will select a subset of COPs relevant for 6G. The project aims at developing a foundation model to tackle the selected COPs, combining GNNs and LLMs to serve as basis for intent-based optimization.

Planned approach and implementation

The thesis project will have the following structure: - Select subset of COPs relevant for 6G resource allocation. - Litearture review of GNNs or LLMs applied to COPs. - Design a model architecture combining GNNs and LLMs for intent-based optimization, as well as the necessary data and training procedure. - Experimentation and validation of designed model. All code, datasets and pre-trained models developed during the thesis will be publicly available.