Graph Deep Learning for Wireless Communications: Principles, Applications, and Open Challenges

#communications #information-technology #graph-neural-networks #IEEE #ComSoc #machine-learning

Abstract

As wireless networks become increasingly dense and heterogeneous, managing interference and allocating resources demand methods that can account for complex interactions among network entities. Graph deep learning (GDL) provides a natural framework for learning from these interactions by incorporating network structure into neural architectures.

This talk introduces the principles of GDL and explains why the structural properties of graph neural networks, including permutation equivariance and local message passing, make them well suited to wireless communication problems.

Drawing on examples including space-air-ground integrated networks and cell-free massive MIMO, it illustrates how graph representations can support resource allocation and network optimization, including through integration with deep reinforcement learning.

The talk concludes with open challenges in generalization, distributed implementation, and robustness under realistic wireless conditions, highlighting opportunities for further research at the intersection of wireless communications and machine learning.

Speaker Biography

Peng Cheng is an Associate Professor in the Department of Computer Science and Information Technology at La Trobe University, Australia.

His research interests include machine learning, graph neural networks, and their applications in wireless communications and intelligent networks.