Im Wintersemester 2024/25 findet das Institutsseminar nach Bedarf mittwochs 14:30 Uhr im Raum PRÜ-1104 statt.
| Datum | Thema | Vortragende/Vortragender |
|---|---|---|
| 23.10.2024 | Python Environments & Docker This talk will provide an introduction to Python package management including some best practices for conflict resolution and the effective use of Python virtual environments. We also give an introduction to docker and singularity. Topics Covered: Environments (Python vs. System), Tools (pip, [Ana]conda, ...), Resolving Conflicts, Python Virtual Environments, Introduction to containerization with Docker and Singularity | Matthias Werner |
| 06.11.2024 | Surrogate-Accelerated Markov Chain Monte Carlo Methods for Bayesian Inverse Problems This talk focuses on efficient sampling using Markov chain Monte Carlo (MCMC) methods for Bayesian inversion. The sampling process is based on the Delayed Acceptance Metropolis-Hastings algorithm and accelerated through adaptively constructed surrogate models, while maintaining the asymptotic exactness of the Markov chain. The surrogate models are built using non-intrusive methods; specifically, neural network surrogate models will be discussed. The motivation for the use of the Bayesian approach stems from the need to solve geotechnical inverse problems with uncertainties, such as identifying material parameters based on data from a tunnel sealing experiment or determining fracture apertures. However, the methods are also applicable beyond geosciences. In considered problems, the forward model typically involves solving partial differential equations using numerical methods. Using surrogate-accelerated MCMC methods, the number of required forward model evaluations can be significantly reduced. The resulting MCMC framework is available in the form of a Python package, SurrDAMH (github.com/dom0015/surrDAMH). | Simona Beresova, PhD (Institute of Geonics) |
| 27.11.2024 | Bayesian inference of covariate-parameter relationships for population modelling An important goal in pharmacology is to tailor drug doses to each patient. To this end, one often uses parametrised ODE initial value problems to model the time evolution of drug concentrations in the body after administration. The parameters in these models often cannot be measured directly in clinical settings, and per-patient data may be too sparse to permit reliable parameter inference for each patient. One approach to this problem is to identify a set of covariates that are clinically measurable, e.g. age and weight, and to specify a covariate-parameter relationship, i.e. a function that maps every admissible covariate vector to a parameter vector. This establishes a nonlinear regression problem, where the i-th covariate Xi is the vector of covariates for the i-th patient, the i-th response Yi is a vector of blood drug concentrations collected at finitely many times for the i-th patient, and the forward model depends on an unknown covariate-parameter relationship. The task is then to find the most appropriate covariate-parameter relationship from some admissible class. We show how this task can be tackled for a family of parametrised ODEs, by using a framework for Bayesian nonlinear statistical inverse problems developed by Nickl et al., to show posterior contraction and a Bernstein-von Mises result. | Prof. Han Cheng Lie (University of Potsdam) |
| 11.12.2024 | Ein nichtparametrischer Algorithmus zur Erkennung von Wechselpunkten (Verteidigung Masterarbeit, online via Zoom) | Mohamed Reda Amarray |
| 18.12.2024 | Automated Identification of Anomalies in Technical Assets of Trains (Verteidigung Masterarbeit) | Jerald Reventh Irudayraj |
| 08.01.2024 | ||
| 15.01.2024 | ||
| 22.01.2024 | ||
| 29.01.2024 | ||
| 05.02.2024 |
Vergangene Vorträge
Im Sommersemester 2024 haben wir uns mit dem Thema Kausalität auseinandergesetzt und dabei wesentliche Teile des Buches "Elements of Causal Inference" von Jonas Peters, Dominik Janzing und Berhard Schölkopf (MIT Press, 2017) besprochen:
| Datum | Thema | Vortragende/Vortragender |
|---|---|---|
| 17.01.2024 | Zu einer Frage der bedingten Abhängigkeit | Prof. Hans-Jörg Starkloff |
| 24.04.2024 | Statistik und Kausalität | Prof. Hans-Jörg Starkloff |
| 08.05.2024 | Einführung in kausale Inferenz (Abschnitt 1 und 2) | Prof. Björn Sprungk |
| 15.05.2024 | Strukturelle kausale Modelle (SKM) und Interventionen (Abschnitt 6.1 - 6.3) | Dr. Christoph Brause |
| 29.05.2024 | Kontrafaktizität, Markoweigenschaft, Treue und kausale Minimalität (Abschnitt 6.4 - 6.5) | Kevin Bitterlich |
| 05.06.2024 | Berechnung von Interventionsverteilungen, Do-Kalkül und Falsifizierbarkeit (Abschnitt 6.6 - 6.8) | Prof. Björn Sprungk |
| 12.06.2024 | Kausale Inferenz mit Python (ZIP-Datei mit Code und Folien: hier) | Matthias Werner |
| 19.06.2024 | Mögliche Ergebnisse und strukturelle Identifizierbarkeit in SKM (Abschnitt 6.9 und 7.1) | Konstantin Ibadullaev |
| 26.06.2024 | Methoden zur strukurellen Idenfikation in SKM (Abschnitt 7.2) | Hanyue Gu |
| 03.07.2024 | SKM und Verborgene Variablen (Abschnitt 9.1 bis 9.4) | Dr. Anna Chekhanova |
| 10.07.2024 | SKM für Zeitreihen (Abschnitt 10) | Dr. Andreas Wünsche |
- 08.12.2023
Jun.-Prof. Daniel Walter (HU Berlin)
Source estimation and optimal sensor placement in spaces of measures - 16.11.2023
Jannis Chemseddine (TU Berlin)
When does minimizing the joint error yield good posterior reconstructions? - 16.11.2023
Paul Hagemann (TU Berlin)
Stability of Conditional Generative Models w.r.t. Observations - 20.06.2023
Dr. Orkun Furat (U Ulm)
Virtual materials testing: Workflow from image processing, via stochastic modeling to numerical simulation for establishing structure-property relationships - 26.05.2023
Prof. Marcus Wiens (TU BA Freiberg)
Measures for criticality and useful redundancy for the design of robust supply networks
- 06.12.2022
Prof. Rudolf Kruse (U Magdeburg)
Unsicheres Wissen in der Künstlichen Intelligenz: Methoden und Anwendungen - 09.11.2022
Prof. Dietrich Stoyan (TU BA Freiberg)
Über den vermeintlichen Beweis der Markt-Hypothese zum Ursprung der Covid-19-Pandemie - 28.07.2022
Prof. Amir Sagiv (Columbia University)
A Measure Perspective on Uncertainty Propagation - 13.07.2022
Prof. Kengo Kamatani (ISM Osaka)
Non-reversible guided Metropolis kernel - 29.06.2022
Jun.-Prof. Conrad Jackisch (TU BA Freiberg)
Unsicherheiten in freier Wildbahn – Erfahrungen aus einem interdisziplinären Forschungsprojekt zu Anpassung an den Klimawandel an der Nordseeküste - 14.06.2022
Prof. Daniel Rudolf (U Passau)
Slice Sampling - 01.06.2022
Prof. Dietrich Stoyan (TU BA Freiberg)
Anpassungstests für Daten aus Siebanalysen - 11.05.2022
Prof. Sebastian Schmon (U Durham)
Optimal scaling of random-walk Metropolis algorithms using Bayesian large-sample asymptotics - 15.02.2022
Kevin Bitterlich (TU BA Freiberg)
Nichtreversible Metropolis-Hastings Algorithmen - 01.02.2022
Prof. Hans-Jörg Starkloff (TU BA Freiberg)
Zur Existenz von Lösungen von zufälligen gewöhnlichen Differentialgleichungen in Skalen von Banachräumen - 18.01.2022
Jun.-Prof. Björn Sprungk (TU BA Freiberg)
Stability of Uncertainty Quantification and Bayesian Inverse Problems
- 30.11.2021
Sebastiano Grazzi (TU Delft)
The Boomerang sampler - 16.11.2021
Prof. Dietrich Stoyan (TU BA Freiberg)
Punktprozess-Statistik in der Partikelgrößen-Statistik - 02.11.2021
Dr. Philipp Wacker (FAU Erlangen-Nürnberg)
Deterministic Dynamics of Ensemble Kalman Inversion - 19.10.2021
Paula Klinger (TU BA Freiberg)
Monte Carlo-Berechnung von Risikomaßen mit Anwendung in der Energiewirtschaft - 13.07.2021
Viacheslav Natarovskii (GAU Göttingen)
Slice Sampling - 15.06.2021
Dr. Simon Weissmann (U Heidelberg)
Analysis of the ensemble Kalman inversion: from discrete to continuous time - 01.06.2021
Markus Dietz (TU BA Freiberg)
On a stochastic arc furnace model - 18.05.2021
Prof. Dr. Han Cheng Lie (U Potsdam)
Stochastic optimal control of SDEs for rare events and importance sampling of path functionals - 04.05.2021
Dr. Jonas Latz (U Cambridge)
Analysis of Stochastic Gradient Descent in Continuous Time - 20.04.2021
Dr. Jeremy Budd (TU Delft)
Theory and Image Segmentation with Graph MBO and Allen-Cahn