More about HKUST
A Survey on the Evaluation of Federated Learning
PhD Qualifying Examination Title: "A Survey on the Evaluation of Federated Learning" by Mr. Di CHAI Abstract: Evaluation is a systematic method to study how well a system achieves its goal. Federated learning (FL) is a new paradigm of privacy-preserving machine learning, which enables different parties to jointly train models without exchanging private data. The evaluation of FL is challenging because it is an interdisciplinary area and has many goals, e.g., privacy, security, efficiency, and heterogeneity. Meanwhile, the evaluation of FL is essential and has many applications. Firstly, FL is an application-driven technology, and the evaluation works as access control of FL methods in the application such that methods with severe issues are filtered out in the application. Secondly, the evaluation system works as an incentive mechanism to evaluate each party's contribution and do a fair payoff sharing. Thirdly, the evaluation system can work as an online and life-long verification to enhance the FL methods' security and privacy. Since most of the FL studies assume the participants are semi-honest, which cannot be guaranteed in real-world applications. A real-time and life-long evaluation is essential to detect malicious behaviors and guarantees that the participants strictly follow the predefined secure protocol. In this survey, we will first investigate the evaluation aspects used in existing works and categorize the evaluation metrics. Afterward, we will look into each evaluation aspect and introduce the evaluation approaches used in each aspect. Finally, we give a set of challenges and future research directions for the evaluation of FL. Date: Thursday, 30 September 2021 Time: 2:00pm - 4:00pm Zoom meeting: https://hkust.zoom.us/j/92265505543?pwd=cnFFY3AxWDNEeXl0WUh3cC9pWEsxdz09 Committee Members: Prof. Qiang Yang (Supervisor) Dr. Kai Chen (Supervisor) Dr. Qifeng Chen (Chairperson) Prof. Ke Yi **** ALL are Welcome ****