Disinformation as Data Poisoning

Principal Investigator:

Daniel Runfola, assistant professor, applied sciences, College of William & Mary

Co-principal investigator:

Anthony Stefanidis, professor, computer science, William & Mary

Disinformation as Data Poisoning

The William & Mary geoLab has hosted two fellowship projects with the Commonwealth Cyber Initiative (CCI), engaging 99 students in projects exploring the intersection of deep learning, data poisoning, and satellite imagery. 

These projects have been largely implemented in collaboration with defense and intelligence industry partners, and have led directly to internship and job opportunities.  

Researchers are proposing a third round for this project, to focus on disinformation as data poisoning. In this version, 25 additional students will work closely with existing and new defense and intelligence partners to test and prototype techniques to identify and automatically mitigate data poisoning in social media streams.  

The work will build on the fundamental hypothesis that techniques that are effective in detecting data poisoning in imagery models (i.e., corrupting pixels in an image to distort deep learning models outputs) could also be substantially helpful in detecting data poisoning in models integrating social media (i.e., disinformation that’s ‘poisoning’ the corpus of collected tweets). 

Existing projects have been both cost-effective and scalable, enabling researchers to build substantial local infrastructure to support increasingly lower-cost student engagement with project partners and advisors.

This project, among nine experiential learning projects funded for fiscal year 2022-23, is a part of CCI's strategy to provide students with industry experience and enhance their skill sets to better prepare them to enter the cybersecurity workforce.