Speaker: Jamie Hayes, UCL-CS
UCL Contact: Vasilios Mavroudis (Visitors from outside UCL please email in advance).
Date/Time: 18 Jan 18, 16:00 - 17:00
Venue: New Quad Pop Up 102
This talk gives an overview of his two recent papers on use cases of adversarial machine learning. The first half presents the first ever method for information hiding using machine learning that is competitive with more established techniques. The second half presents a case study into privacy leakage of generative models. With no knowledge of the training algorithm, model type or architecture, model parameters, or data distribution, we show an attacker can learn which data items were used to train a generative model, which is potentially privacy sensitive information.
Jamie Hayes is a PhD student from University College London, UK. His research focuses on the intersection of privacy, security and machine learning.