Robust and high performance machine learning for next generation wireless networks

Liu, Zikun

Permalink

https://hdl.handle.net/2142/129421 Copy

Description

Title

Robust and high performance machine learning for next generation wireless networks

Author(s)

Liu, Zikun

Issue Date

2025-04-23

Director of Research (if dissertation) or Advisor (if thesis)

Vasisht, Deepak

Doctoral Committee Chair(s)

Vasisht, Deepak

Committee Member(s)

• Choudhury, Romit Roy
• Caesar, Matthew
• Singh, Gagandeep
• Xie, Yaxiong

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine Learning
• Wireless Networks
• 4G
• 5G
• Next Generation Networks
• Robust Training
• Adversarial Attack
• Satellite Network
• Time Series Prediction
• Sustainability
• AI
• Articial Intelligence
• Wireless Communication
• Generative Networks
• Transformers
• GAN
• Variational Autoencoders
• LSTM
• LLM

Abstract

Next-generation (NextG) wireless networks promise unprecedented scale, heterogeneity, and capabilities, enabled by rapid advancements in networking hardware such as large-scale antenna arrays, low-earth-orbit (LEO) satellite systems, and edge-computing infrastructures. However, realizing the full potential of these technologies requires addressing significant challenges in system optimization, adaptability, and robustness under real-world constraints. This thesis presents a set of machine learning-based approaches that co-design algorithmic intelligence with emerging wireless hardware to optimize network performance, reliability, and robustness. First, this work designs machine learning-based systems to adaptively leverage new wireless infrastructures, such as satellite networks and 4G/5G base stations, to maximize throughput and reduce communication overhead in volatile environments. These systems incorporate domain-specific insights and predictive modeling to optimize resource allocation and network behavior in real time. Second, this work exposes critical vulnerabilities in existing wireless ML systems by designing practical adversarial attacks that survive over-the-air distortions. We show that these attacks can degrade performance in real deployments, motivating the need for fundamentally more robust solutions. Finally, this thesis discusses future directions, including provably robust learning for wireless systems, energy-efficient techniques to reduce the carbon footprint of NextG infrastructure. By bridging the gap between cutting-edge machine learning and real-world wireless systems, this work contributes toward building robust, efficient, and adaptive networks for the next generation of connectivity

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129421

Copyright and License Information

Copyright 2025 Zikun Liu

Owning Collections