Ziel der VL ist es ein grundlegendes Verständnis von KI-Systemen/Algorithmen mit Datenschutz.

(English) Outline (tentative):

1.    Data bases: Nomenclature, data representations, queries and statistics, attacks
2.    DeepL I: Topology, universality, stochastic gradient
3.    DeepL II: Generative modeling, variational inference, VAE, ELBOW
4.    DeepL III: GAN architecture, metrics, convergence
5.    DeepL IV: Variational autoencoder famility, information bottleneck, dual formulation, limits and tradeoffs
6.    Diff. Privacy I: Definitionen, Gaussian DP, DP as hyphothesis tests
7.    Diff. Privacy II: Mechanisms (Gaussian and Laplace), compositions
8.    Diff. Privacy III: Performance limits und tradeoffs
9.    DeepL mit Diff.Privacy I: DP-SGD
10.    DeepL mit Diff.Privacy II: PATE
11.    Synthetic Data I
12.    Synthetic Data II
13.    Robustheit/Certification I
14.    Robustheit/Certification II
15.    Applications I: Biometrics
16.    Applications II: tba