Data Poisoning and Satellite Reconnaissance: Bridging Application and Education
Daniel Runfola, assistant professor of computer science; Anthony Stefanidis, professor of computer science; and Peter Kemper, associate professor of computer science.
Convolutional neural networks have been extensively used to assess road networks using imagery collected in two ways: on the ground via smartphone and aerially via satellites or drones. Imagery collected on the ground is largely used for road quality or as input into self-driving algorithms, whereas imagery collected by satellite or airborne systems is largely used to identify the growth or change in road networks in inaccessible areas. In this project, students will work with William & Mary faculty and external industry partners to design a data collection app and construct a convolutional neural network to predict road roughness across Virginia, incorporating data from ground and aerial imagery. The second part of the project will be a cybersecurity competition in which student teams will explore ways to identify potential malicious changes to imagery and ways to mitigate the impact of such “data poisoning.”