Federated learning is a paradigm breakthrough shift in AI for data privacy preservation. Unlike conventional AI techniques, federated learning does not require the users to share the data, but only the AI model parameters. As such, federated learning has found notable successes in numerous applications (e.g., Google keyboard, localization, security, and data sharing) and various disciplines (e.g., finance, healthcare, and mobile networking). However, the implementation of federated learning algorithms requires considering both computing and communication aspects, in which the communication between the learning server and FL clients is over wireless networks (e.g., beyond 5G and 6G). I can supervise the following topics of wireless federated learning.
- Communication-efficient federated learning: Communication efficiency is an utmost important issue in distributed machine learning approaches like federated learning. There are many ways to design communication-efficient federated learning, such as AI model compression and communication resource management. This project aims to design a new communication-efficient approach to federated learning. The primary objective of this topic is to understand the communication efficiency issue in federated learning. The key goals are as follows:
- Understand the concept of distributed learning and federated learning.
- Understand the communication efficiency issue in federated learning.
- Apply engineering mathematics and convex optimization to analyze the model convergence and then optimize the considered federated learning system.
- Federated learning over cybersecurity: Federated learning aims to protect data privacy and somehow enhance user security. However, cyberattacks are frequently happening in various services and applications that are deployed in real-time and over wireless communication systems. This topic aims to develop new federated learning methods to enhance data privacy and improve the performance of cyber systems. The primary objective of this topic is to understand and explore how federated learning can improve the performance of cybersecurity systems. The key goals are as follows:
- Understand the concept of distributed learning and federated learning.
- Understand the fundamentals of cybersecurity and explore important research aspects of cybersecurity in protecting wireless networks and IoT.
- Investigate new deep federated learning methods for IoT cyberattack classification.
Federated unlearning (and machine unlearning) is a key privacy-enhancing technology to enable data erasure and the right to be forgotten for everyone in the digital era. More broadly, it supports the implementation of AI legislation and data protection regulations, such as the EU AI Act and the GDPR. The primary objective of this topic is to understand how unlearning can be designed and performed efficiently. The key goals are as follows:
- Design new incentivisation methods to motivate users in participating in the unlearning process.
- Design new methods to ensure that unlearning can be readily and verifiably performed.
- Evaluate learning and unlearning performance across various evaluation scenarios and data modalities.
In addition to the topics mentioned above, I am happy to discuss any project ideas (e.g., federated unlearning and decentralized learning) about this exciting research area. More information: https://www.scss.tcd.ie/viet.pham