Objective
This project aims to develop privacy-preserving deep learning algorithms, including the design of architecture, federated learning, inference mechanism, privacy protection procedure, and apply them into practical uses. 

The project is divided into three parts. In the first part, we aim to deal with the privacy leakage caused by deep learning models, as they often extract redundant sensitive attributes that aren't necessary to the main task. The second part aims to serve the deep learning requests from today's smart cities/homes. The third part aims to develop federated reinforcement learning algorithms applied to offloading optimization problems in mobile edge computing networks.