Upcoming events

25 November 2024 15:45 till 16:45

[STAT/AP] Julia Olkhovskaya: TBA

26 November 2024 12:30 till 13:45

[AN] Michael Kniely: Thermodynamically Correct Models for Electro-Energy-Reaction-Diffusion Systems

We introduce a thermodynamically consistent framework for reaction-diffusion systems modeling the evolution of a finite number of charged species. This approach covers, in particular, a large class of inorganic semiconductor-type models. Here, thermodynamical consistency refers to charge and energy conservation and the production of entropy. This is achieved by formulating the model as a gradient flow system in Onsager form (for the concentrations and the internal energy) coupled to Poisson's equation (for the electrostatic potential).

First, we will focus on the structure of the Onsager operator and its relation to the electrostatic potential. Similarities and differences to other temperature-dependent semiconductor-type models shall be discussed as well. Another aspect is the well-posedness of the corresponding equilibrium problem. By resorting to the Lagrange formalism, one can rewrite this entropy maximization problem under the constraints of charge and energy conservation as a convex minimization problem. We will see that this problem admits a unique solution, hence, a unique thermodynamic equilibrium exists. This project is joint work with Katharina Hopf and Alexander Mielke.

28 November 2024 16:00 till 17:00

[PDE&A] Ivan Langella

02 December 2024 15:45 till 16:45

[STAT/AP] Sonja Cox: TBA

TBA

03 December 2024 16:00 till 17:00

[AN] Francesca Arici: TBA

TBA

10 December 2024 16:00 till 17:00

[AN] Pascal Auscher

TBA

11 December 2024 16:00 till 17:00

[DMO] Martin Scotti: Intersecting codes in the Hamming and in the rank metric

12 December 2024 16:00 till 17:00

[PDE&A] Anh Khoa Doan

13 December 2024 12:30 till 13:15

[NA] Andrea Brugnoli: TBA

TBA

16 December 2024 15:45 till 16:45

[STAT/AP] Frank Röttger: Graphical models in extremes from threshold exceedances

The recent introduction of conditional independence for multivariate extremes from threshold exceedances has inspired a new line of research in extremal dependence modeling. In this talk we summarize recent developments and try to highlight connections with related fields. In particular we discuss undirected and directed graphical models for multivariate extremes from threshold exceedances, as well as approaches for structure and parameter learning. Here, a central tool is the parametric family of Hüsler--Reiss distributions, which can be characterized via Laplacian-constrained Gaussian graphical models, i.e. degenerate Gaussians with graph Laplacian inverse covariance matrix. This characterization allows to describe extremal conditional independence parametrically and therefore leads to a parametric encoding of extremal graphical models. Finally, we introduce colored Hüsler--Reiss graphical models and discuss statistical methodology for those, which we demonstrate on a real data example.

17 December 2024 16:00 till 17:00

[AN] Jan van Neerven: Time

TBA

07 January 2025 16:00 till 17:00

[AN] Tobias Werner: TBA

TBA

13 January 2025 15:45 till 16:45

[STAT/AP] Rianne de Heide: Recent developments in e-value-based multiple testing

16 January 2025 16:00 till 17:00

[PDE&A] Jason Frank

17 January 2025 12:30 till 13:15

[NA] Richard Löscher: TBA

TBA

21 January 2025 16:00 till 17:00

[AN] Esmee Theewis

TBA

30 January 2025 16:00 till 17:00

[PDE&A] Jana Weber

11 February 2025 16:00 till 17:00

[AN] Mark van den Bosch: TBA

TBA

13 February 2025 16:00 till 17:00

[PDE&A] Rishabh Gvalani

14 February 2025 12:30 till 13:15

[NA] Amanda Howard: Multifidelity, domain decomposition, and stacking for improving training for physics-informed networks

Physics-informed neural networks and operator networks have shown promise for effectively solving equations modeling physical systems. However, these networks can be difficult or impossible to train accurately for some systems of equations. One way to improve training is through the use of a small amount of data, however, such data is expensive to produce. We will introduce our novel multifidelity framework for stacking physics-informed neural networks and operator networks that facilitates training by progressively reducing the errors in our predictions for when no data is available. In stacking networks, we successively build a chain of networks, where the output at one step can act as a low-fidelity input for training the next step, gradually increasing the expressivity of the learned model. We will finally discuss the extension to domain decomposition using the finite basis method, including applications to newly-developed Kolmogorov-Arnold Networks.