Seminar | November 22 | 3:10-5 p.m. | 107 South Hall
Ankit Bansal & Mugdha Bhusary
Two students present their final progress reports:
Protecting World Leaders Against Deepfakes
The creation of sophisticated fake videos has been largely relegated to Hollywood studios or state actors. However, advances in deep learning have made it very easy to create sophisticated fake videos called deep fakes. These pose a significant threat to democracy, national security, and society. Researchers have developed forensic techniques to typify an individualâs speaking pattern. However, these methods need to scale. So I am working on a technique to automate the creation of deepfakes in order to provide a better test environment for forensic technique developers. I will also be addressing the current challenges in the creation of these videos
Managing Pedestrian Privacy in an Age of Self-Driving Cars
Self driving or assisted driving cars are equipped with 360Â cameras. They are constantly recording people and license plates around them. However using the protected information of peopleâs faces and license plates for a task is not acceptable. Through the time of this course, a project has been developed that would look at the potential of differential privacy and adversarial machine learning to tackle this problem. It would evaluate the possibility of adding a perturbation (measured noise) while ensuring model accuracy is not affected. Further, once we realize the potential of building such a model, the project would tackle the question of where and how the notification should be implemented.