Title: Few-Shot Knowledge-Enhanced Federated Learning
Department: Computer Science
Description: The main purpose of this project is to improve the prediction accuracy as well as the privacy and security for Federated learning (FL) models. Federated learning is an effective privacy-preserving mechanism that collaboratively trains a shared model on decentralized data without data exchange. It has shown great potential to protect the data privacy of users, which makes it widely used in a variety of fields including healthcare, the Internet of Things (IoT), and autonomous driving. However, existing FL models only make predictions based on the data-driven method while neglecting domain knowledge. In real-world applications like AI disease diagnosis, domain knowledge from experts plays a vital role in modeling, which can largely enhance prediction accuracy. On the other hand, most existing FL methods require a large number of communication rounds between the local clients and the central server for transmitting model parameters. This leads to communication overhead, safety, and security issues. Therefore, this project will focus on resolving these two major issues: incorporate domain knowledge into FL to improve prediction accuracy and reduce the communication rounds for protecting model privacy and security.
Hometown: Hangzhou, China
Advisor: Huajie Shao
Each William & Mary Honors Fellowships donation supports the whole Charles Center Honors Fellowships Fund. Donations inspired by specific research projects are distributed by the Charles Center in a way that benefits this project and others. To learn more about Honors Fellowships, please visit our website.