5/5 Our work provides a novel class of simple counterattacks against deep-fake detection. The attacks can be used as baseline to evaluate the robustness of frequency-based deep-fake detectors.
1/5 I will present a novel counterattack against deep-fake detection methods at the DLS workshop @IEEESSP today. Together with @mlsec @darpsec @ErwQui, we manipulated the frequency domain of deep fakes to evade detection. Link: https://t.co/s4HVs6ki9a
4/5 We evaluate our attacks on three different classifiers that use the frequency spectrum; one fingerprint-based, and two learning based. The detection rate from each GAN can be reduced by one of our attacks.
Manipulierte #Deepfakes, Lernalgorithmen für #Roboter und Neugier für bessere #KI: Auf der #TECHTIDE in #Hannover sind in der vergangenen Woche drei Nachwuchswissenschaftler*innen der #TUBraunschweig für ihre Abschlussarbeiten ausgezeichnet worden.🏆🥳
➡️https://t.co/j4DvaeepFe
1/ I presented the paper "In-Depth Technical and Legal Analysis of Tracking on Health Related Websites with Ernie Extension" @ WPES'21. A pleasure to work with @Fouad__Imane, @Cristianapt, @Yanis_Boussad, @nataliabielova, @arnaudlegout. Thread ⬇️. Link: https://t.co/3DpmeJVnFU
8/ When the cookies performing cross-site tracking are related to 𝗮𝗱𝘃𝗲𝗿𝘁𝗶𝘀𝗶𝗻𝗴 𝗽𝘂𝗿𝗽𝗼𝘀𝗲𝘀, it raises privacy concerns, since it is possible to build and enrich unique user profiles based on sensitive health data. We demonstrate this issue on 3 case studies.