Michael Fauss - Dipl.-Ing. degree from Technische Universität München Communication systems
Date: - Location: Eurecom
Abstract: Robust statistics continue to gain importance in practical engineering systems. Common issues that can be addressed by robust methods are, for example, outliers in the data or mismatches between model and reality. The talk aims at discussing some recent efforts in this area. First, a brief review of the minimax design principle is given, and nonparametric neighborhoods are introduced to capture distributional uncertainties. In the second part, the design of minimax optimal detectors is discussed for neighborhoods of the f-divergence-ball type and the density-band type. Both detectors are shown to be closely related and to offer a great degree of flexibility while remaining mathematically tractable. In the last part of the talk, a recent result on MMSE estimation for neighborhoods of the relative-entropy-ball type is presented and potential applications in robust distributed estimation are discussed. Bio: Michael Fauss received the Dipl.-Ing. degree from Technische Universität München, Germany, in 2010 and the Dr.-Ing. degree from Technische Universität Darmstadt, Germany, in 2016, both in electrical engineering. In November 2011, he joined the Signal Processing Group at Technische Universität Darmstadt, where he is currently working as a research fellow. In 2017 he received the dissertation award of the German Information Technology Society for his PhD thesis on robust sequential detection. As of September 2019 he will join Prof. H. Vincent Poor's group at Princeton University as a postdoc on a two-year research grant by the German Research Foundation (DFG). His current research interests include statistical robustness, sequential detection and estimation, and the role of similarity measures in statistical inference.
