Thesis Talk Mick Tuit – Image Protection Against DeepFakes Using Adversarial Attacks

Deepfake Detection is the task of deciding whether a video is a deepfake or not. With the increase of popularity in deepfakes and its malicious applications, the interest in ways to detect these deepfakes has also grown.

The reason I chose this subject for my thesis is a combination of three things. Firstly, I have a great interest in the technologies which are involved in creating deepfakes, i.e., Deep Learning. Secondly, the (potential) impact of deepfakes on society and how we can prevent it is something that also resonated with me. I feel it is important to be working on societal problems within AI and Deep Learning. Lastly, the subject is currently heavily researched and thus the research is highly relevant and can potentially be very impactful on the area. All in all, I expect deepfakes are here to stay and we need to be able to combat them whether it through detection or perhaps regulations.