Training native intelligent communication systems
File(s)
Author(s)
Kaidi, Xu
Type
Thesis
Abstract
Native intelligence, the paradigm from embedding to bonding artificial intelligence with wireless communication systems together, plays an important role in next-generation communication systems to render scalable, low-latency, and high-efficiency services. However, due to the complex nature of communication systems, such as time-varying wireless channel statistics, user mobility, and massive connections, and the distributed data across the communication networks, which leads to local observability, the widely used training methods from the machine learning community often fail to efficiently train multiple intelligent agents for communication systems. Therefore, in this thesis, we have proposed a series of advanced distributed training frameworks to efficiently train the native intelligent networks from four different perspectives.
Version
Open Access
Date Issued
2025-06-28
Date Awarded
01/12/2025
License URL
Advisor
Geoffrey, Li
Publisher Department
Department of Electrical and Electronic Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)