CyberSec Seminar October 2022
For this month's seminar, our speakers are:
- Max Noppel (Intelligent System Security): Advanced Problems in Backdooring Explainable Machine Learning
Abstract: Explainable machine learning should help us understand and validate the decision making of a ML model. While shallow decision trees are explainable by default, more complex models need additional algorithms to reach this goal. These usually assign relevance values per input feature and can be visualizied as heatmaps, highlighting where the model "looks". Unfortunately, these explanations can be manipulated through input manipulations or model manipulations. In this talk, we focus on a backdoor injection attack for explanations, a specific form of model manipulation. After introducing the attack in an strong attack model we discuss advanced problems in this regard.
- Marcel Tiepelt (Cryptography and Security): Quantum Enumeration -- How to Find Children Hiding in Random Trees
Abstract: Post-quantum cryptography will be the de facto standard for secure communication in the near future. With an adversarial model (=quantum computers)
that does not exist in practice yet, the amount of crypt-analysis that goes into finding weaknesses is limited. For now, the best we can do is to estimate
the cost of an attack using heuristics, and adjust key-size to a level where we are confident that no (known) quantum algorithm can break the schemes.
In this talk, we want to take a look at analyzing lattice based cryptography in the setting of quantum enumeration.
After the seminar, we will head over to Pizzeria Il Caminetto for the Security & Privacy Lunch.