Research Group "Stochastic Algorithms and Nonparametric Statistics"

Seminar "Modern Methods in Applied Stochastics and Nonparametric Statistics" Summer Semester 2024

16.04.2024 Ganet Some Maalvladedon (African Institute for Mathematical Sciences-Ghana)
Stochastic optimal control of a prosumer in a district heating system
We consider networks of residential heating systems in which several prosumers satisfy their heating and hot water demand using solar thermal collectors and services of a central producer. Overproduction of heat can either be stored in a local thermal storage or sold to the network. Our focus is the minimization of the prosumers expected total cost from purchasing and selling thermal energy and running the system. This decision making problem under uncertainty about the future production and consumption of thermal energy is formulated as a stochastic optimal control problem and solved with dynamic programming techniques. We present numerical results for the value function and the optimal control.
23.04.2024 Priv.-Doz. Dr. John Schoenmakers (WIAS Berlin)
Optimal stopping with randomly arriving opportunities to stop
30.04.2024

07.05.2024

14.05.2024 Dr. Thomas Möllenhoff (RIKEN, Japan)
Variational inference with natural gradient descent
21.05.2024
Different Room: ESH
28.05.2024

04.06.2024

11.06.2024 Dr. Gabriele Iommazzo (Zuse Institute Berlin)
Starting at 4 pm! Linearly converging conditional gradients over intersecting polytopes
In this work, we present a variant of the Frank-Wolfe algorithm designed to efficiently find a point in the intersection of two or more polytopes. We achieve this by optimizing a function that penalizes the distance between variables in each polytope and by computing updates within the Cartesian product of the sets. Although the objective function is not strongly convex, we establish linear convergence rates by deriving bounds on the product domain's pyramidal width, which we relate to the pyramidal widths of each individual polytope. The proposed algorithm not only addresses the approximate feasibility problem over the intersection of polytopes, but can also be used to efficiently approximate linear programming problems over the same feasible region by treating the LMOs of each polytope as black-box operations. We provide empirical evidence showcasing the applicability of the algorithm across large-scale optimization tasks.
18.06.2024

25.06.2024

02.07.2024

09.07.2024

16.07.2024

23.07.2024

30.07.2024



last reviewed: May 31, 2024 by Christine Schneider