Project 42: Design and Implementation Scheme of Privacy Protection in AI-Enabled Internet of Things
Contact Information
Prof. Yue Wu
Email: wuyue@sjtu.edu.cn
Project Description and Objectives
Considering the image transmission in an AIoT scenario, the private and useful information are always entangled and hard to split by simple processes in data space, such as cropping a certain region of pixels. Traditional privacy protection methods often use differential privacy to add noise to the entire data, which not only causes a huge waste of resources and computing power of edge devices, but also adversely affects the performance of downstream tasks. By proposing and utilizing a feature decoupling algorithm, the private features and features corresponding to downstream tasks are extracted separately. Also, selective noise addition can be achieved. This project intends to realize targeted privacy protection by identifying the private features and applying differential privacy to protect users’ private data in AIoT conditions.
Eligibility Requirements
Basic knowledge of machine learning, matrix analysis, and optimization theory.
Mastery of Python and at least one deep learning framework, Pytorch preferred.
Main Tasks
Review literature on differential privacy.
Achieve a prototype of a feature decoupling algorithm to extract privacy-related features.
Website
Lab: http://nelcat.sjtu.edu.cn/
School: http://infosec.sjtu.edu.cn