Direkt zum Inhalt springen
login.png Login    |
de | en
Technische Universität München
150 Jahre culture of excellence – Besuchen Sie die Jubiläumswebsite www.150.tum.de!

Technische Universität München

Sitemap > Schwarzes Brett > Abschlussarbeiten, Bachelor- und Masterarbeiten > Master Thesis: Adversarial Examples for Neural Networks
auf   Zurück zu  Nachrichten-Bereich    vorhergehendes   Browse in News  nächster    

Master Thesis: Adversarial Examples for Neural Networks

04.01.2019, Abschlussarbeiten, Bachelor- und Masterarbeiten

Adversarial Examples are a fundamental problem observed in Deep Learning, and potentially a huge safety and security issue. Tiny perturbations of an input can lead to completely undesired behaviour of systems that were believed to be accurate. But how big is this threat in reality? What does it mean for Machine Learning theory? And what can we do about it?

Our team is based at the Chair of Robotics, Artificial Intelligence and Real-time Systems at TUM, and at fortiss, which is the the new Bavarian research center for AI. Our work encompasses safety and security for both the practical application of AI systems (e.g. autonomous driving), and also the very fundamentals of Machine Learning. Recently, we won a prize in the prestigious NeurIPS 2018 Adversarial Vision Challenge for our work on black-box adversarial attacks - fooling Neural Networks is a lot of fun!

We are offering several Master's Theses:

If you are interested in this area of research, do not hesitate to contact me!

Kontakt: t.brunner@tum.de