Lecture Series (no. 19331401) + Exercise (no. 19331402)
Foundations of Information-Theoretic Identity and Privacy (Informationstheoretische Grundlagen von Identität und Datenschutz), 10 LP
Schedule: WS, 4+2 SWS
Instructors: Gerhard Wunder, Richard Schöffauer, Benedikt Groß
The lecture series Information-theoretic Foundations of Security (Wintersemester) as well as Privacy and Identity (Sommersemester) provides a comprehensive treatment of the information-theoretic priciples of security and privacy engineering by exploiting (common) randomness vs. classical computational complexity. It is also an excellent guidance for solving statistical decision problems in data science and machine learning.
An intriguing application is "Physical Layer Security", i.e. implementation of security algorithms already on the physical layer of (wireless) communication systems exploiting physical properties such as noise, fading etc. for symmetric key generation and wiretap channel coding ("wiretap codes").
In this lecture series, we will complement the theory of information-theoretic security with a similar framework of information-theoretic founded identity and privacy. We will define relevant key metrics to capture privacy and identity in an information-theoretic framework, which includes typical approaches such as differential privacy in the computer science literature. We shall apply this framework to derive fundamental performance limits of identity provision system, private information retrieval. A dedicated treatment is given to information-theoretic approach to multiparty computation over wireless channels and their related tradeoffs.
Keywords: Information-theoretic privacy and identity management, privacy metrics, differential privacy, fundamental inequalities and achievable performance regions, identity provision systems, private information / function retrieval schemes, privacy-preserving distributed multiparty computation/learning, computation & communication tradeoffs
Seminar (19331611): Information theoretical principles of ML, 5LP
Instructors: Gerhard Wunder, Benedikt Groß
Description: Recently, artificial intelligence and machine learning (AI/ML) has emerged as a valuable tool in the field of communication and signal processing. It is therefore natural to extend the investigations to the field of physical layer security and privacy. This field is still in its infancy with some very preliminary results on wiretap channel code design, feature extraction of wireless channels and a growing part of contributions to privacy-preserving, distributed AI/ML. This seminar will teach the latest advances and synergies between the broad fields of AI/ML and secure communications.
Keywords: ML overview, basic tools, universal approximation, deep learning, stochastic gradient, acceleration strategies, deep convolutional networks, feature extraction, classification, mutual information neural network estimation, structured sparsity in convolutional neural networks, matrix decompositions