Algorithm study of federated learning and its application in radiation treatment planning

Li Ju
Division of Scientific Computing, IT Department, Uppsala University


Abstract:

Data privacy and security have become an obligation for more and more deep learning-based applications in various fields. Federated learning is a collaborative solution for multiple parties to train a joint model without sharing raw data. However, existing federated learning algorithms are yet to be improved for their the risk of privacy leaking, performance deterioration, low communication efficiency, etc. In this presentation, I will do an introduction of my PhD project, which aims to develop algorithms of federated learning and better apply federated learning in radiation treatment planning. Specifically, open problems in federated learning will be overviewed, my current work will be introduced and my future plan will be shared.