Federated Learning over Wireless Networks
Project Description
Federated learning (FL) is envisioned as the bedrock for next-generation intelligent networks to enable distributed AI. In FL, mobile devices collaborate to train machine learning models (i.e., solving the corresponding distributed optimization problems) without uploading all the raw data to a central parameter server for centralized processing. This can avoid high computational burden incurred by centralized learning, and preserve the privacy of mobile users' data, which are often sensitive information. Though being a promising technology, FL faces several technical challenges including:

1. The unreliability of noisy fading wireless channels, which bring distortions to the exchanged information in FL algorithms;
2. The scarce communication resources (e.g., power and bandwidth) of the mobile devices, which limit their capability of exchanging information accurately in the distributed learning algorithms
3. Lack of incentives to participate in FL algorithms, which consume mobile devices' precious radio resources
4. Privacy concerns of the mobile users, as the exchanged information in the distributed learning algorithms may be overhead and used by adversaries to infer the sensitive raw data of users;
5. Security concerns, as malicious entities may send falsified information to disrupt the learning algorithms;
6. Many other issues.
CAO Xuanyu
Course type
Applicant's Roles
The students will conduct cutting-edge research, under my supervision, on FL/distributed leanring over networks, with the goal of publishing high-quality papers on top journals and conferences.

The students must have strong mathematical background, as the research involves lots of mathematics.

Applicants are encouraged to check out my webpage (https://eexcao.people.ust.hk/) to get to know the flavor of my research.
Applicant's Learning Objectives
1. Learning advanced knowledge pertaining to distributed optimization/learning, wireless networks, and other related topics encountered during research (such as game theory/mechanism design, online optimization/learning).

2. Know how to conduct cutting-edge research, including literature survey, looking for a meaningful problem, formulating the problem mathematically, devising algorithms/methods to solve the problem, analyzing the performance of the algorithm, evaluating the algorithm using simulations, writing high-quality papers.

3. Publishing papers on top venues.
Complexity of the project