Project 52: Theory and Practice of Federated Learning
Contact Information:
Prof. Fan Wu
Email: fwu@cs.sjtu.edu.cn
Project Description and Objectives:
The concept of federated learning was first proposed by Google research scientists in 2015, which is a general collaborative machine learning framework for billions of mobile devices with the coordination of the Cloud. Under this framework, (1) a mobile device first downloads a global model from the Cloud, then trains a local personalized model using its user data, and finally uploads the model update to the Cloud, (2) the model updates from multiple mobile clients that are securely aggregated to form a consensus update to the global model in the Cloud, (3) the above process is repeated for the timeliness of the global model and the local models.
In this project, we intend to investigate several theoretical aspects of federated learning, including learning theory, security and privacy, and game theory. We also plan to develop an on-device training framework, and further to deploy our design in practice, benefitting millions of worldwide users.
Eligibility Requirements:
Basic knowledge of machine/deep learning is mandatory.
Basic knowledge of mobile computing, cryptography, and game theory is preferred.
Experience of Android/iOS development is preferred.
Proficiency in writing and speaking in English.
Interest in theoretical analysis or coding.
Main Tasks:
Propose an intriguing idea in the scope of the project.
Present a novel solution and validate its practical feasibility.
Finish a research report in this project.
Website: