Events at Towson University

Mathematics Club Talk - VIRTUAL

Mathematics Club Talk - VIRTUAL

Speaker:    Derek Margulies, Researcher at Noblis

Topic:    Using Generative Adversarial Networks (GANs) to Create Human-Probable Passwords

Abstract:    Abstract: Machine Learning or Artificial Intelligence can be a valuable tool in developing new methods to recover protected information. The ultimate success of these methods is determined by benchmarking the ML approach against existing classical methods, both literally by analyzing data outputs for comparison and also theoretically through detailed analysis of the computational complexity of both methods. Noblis conducted research on a Generative Adversarial Network (GAN) approach to improve the quality, efficiency, and ultimately the rate of success of dictionary-based password attacks on a variety of secure technologies. The algorithm was trained and tested on a dataset of over 33 billion human generated passwords. Using statistical analysis and experimentation on a sufficiently random sample from the exhaustive space, we are able to create a model of human behavior when choosing passwords.

In the GAN approach, we develop a model for password generation which over the course of many iterations of testing and experimentation is refined by an adversarial model, in this case our human-probable classifier. The content generation algorithm thus learns over time what it means to be human-probable, and, if training is successful, will then generate only strings of human-probable passwords.

Dial-In Information

Join Zoom Meeting at

Friday, March 5, 2021 at 2:30pm to 3:00pm

Virtual Event
Event Type

Student Life, Seminar, Virtual


Academic Affairs, Mathematics (Department of), Fisher College of Science and Mathematics

Target Audience


Google Calendar iCal Outlook

Recent Activity