Archive

DAMUT Colloquium archive

2026

  • Date: 01 April 2026  Location: RA-1501
    Speaker:  dr. Frank Röttger (MDS/MOR-STAT)
    Title: An introduction to probabilistic graphical models for multivariate extremes
    Abstract: 
    Multivariate extreme value theory provides a framework for modeling rare events in complex systems, such as floods in river networks, financial crashes, or heatwaves. A central challenge in extremal dependence modeling is to capture the joint structure of extremes using models that remain flexible yet parsimonious, since extreme observations are typically scarce. In this talk, we discuss how probabilistic graphical models address these challenges. We introduce both directed and undirected graphical models for multivariate extremes built from threshold exceedances, and outline methods for structure and parameter learning in high dimensions. Finally, we present a family of extremal graphical models that enables a parametric representationof conditional independence in extremes, leading to particularly tractable and interpretable statistical methodology.
  • Date: 04 March 2026, Location: RA-2501
    Speaker: Dr. Martijn Gösgens (MDS/MOR-SOR)
    Title: Geometry hinders the formation of consensus in asynchronous majority dynamics
    Abstract: 
    We investigate a model of opinion dynamics where vertices asynchronously announce opinions based on their private opinion and on the previously announced opinions of their neighbors. Our focus is on understanding how the structure of the underlying graph influences the likelihood of reaching consensus on the true opinion. Previous work proved that for sufficiently sparse, connected expander graphs, this process terminates in consensus on the true opinion with high probability. In this work, we show that when the underlying graph has geometric structure, the process is likely to terminate in disconsensus. Specifically, we prove that for a one-dimensional Random Geometric Graph (RGG) of n vertices with expected degree o(√n), the process ends in disconsensus with high probability. Numerical experiments indicate that this phenomenon persists in higher-dimensional RGGs. Instead of a global consensus, we observe that the geometry leads to local consensuses.

2025

2024

2023

2022

2021

2020


2019

2018

2017

2016