Advanced Statistical Seminar — NMST611
(Winter Term 2026-2027)
| When: | Wednesdays, 15:40 - 17:00 | (every two weeks) |
| Where: | Lecture Room Praktikum KPMS | (MFF Karlín, Sokolovská 83) |
The Advanced Statistical Seminar consists of presentations delivered, typically in person, by invited foreign speakers and guests of the Department of Probability and Mathematical Statistics at Charles University. The seminar provides a forum for presenting and discussing current developments in modern statistics, covering both theoretical and applied aspects.
A broad range of topics is represented, including methodological advances, new statistical models and techniques, as well as applications motivated by problems from science, economics, finance, and other fields. The seminar also aims to encourage informal discussion and exchange of ideas between invited speakers, members of the department, Ph.D. students, and other participants.
Seminar schedule
- 30.09.2026 | 15:40 | Rainer Dyckerhoff
University of Cologne, Germany
Title: Computing halfspace and simplicial depth in \(O(n^{d-1})\) via topological sweep
We present practical algorithms for the exact computation of halfspace depth and simplicial depth of a single point in \(\mathbb{R}^d\). Both algorithms achieve an asymptotic running time of \(O(n^{d-1})\) . For halfspace depth, this improves upon the previously best-known bound \(O(n^{d-1}\log n)\). In three dimensions, we use a gnomonic projection to map the data to a plane, apply point-line duality, and traverse the resulting line arrangement by a topological sweep. For higher dimensions, we reduce the problem to the three-dimensional case by means of a suitable projection scheme. We also present an exact \(O(n^{d-1})\)-time algorithm for the computation of simplicial depth. This provides a substantial improvement over previous exact approaches. The algorithm is based on the same concepts as that for the halfspace depth. The algorithms combine classical geometric ideas in a practically effective way and, to the best of our knowledge, are the first practically usable algorithms to attain these asymptotic bounds. The talk will emphasize these geometric principles and the implementation techniques that make the algorithms effective in practice.
- 14.10.2026 | 15:40 | Tatyana Krivobokova
University of Vienna, Austria
Title: Hierarchical Lasso for multi-factor designs
In modern experimental sciences, high-throughput experimentation (HTE) generates large datasets with a highly specific structure. For example, in chemical HTE, reaction properties are measured for all combinations of multiple components, forming a factorial design with a single observation per factor-level combination. We show that the model reflecting this data-generating process can be estimated using a hierarchical Lasso estimator that enforces sparsity and hierarchical constraints on main effects and interactions. Then we develop an efficient estimation algorithm, and subsequently show that it achieves consistent variable selection and estimation under mild conditions.
- 11.11.2026 | 15:40 | Sebastiano Vitali
University of Bergamo, Italy
Title: A practical prioritization scheme for goal-based asset-liability management
We present an individual asset-liability management problem characterized, over a potentially long-term horizon, by financial-, consumption- and retirement-type of goals whose achievement probability is maximized under a rich set of constraints in a stochastic linear programming formulation. The decision maker will determine an optimal multistage asset-liability strategy based on a mean-risk paradigm in which the risk is captured by the terminal Conditional Value-at-Risk (CVaR) measure, and the mean by the expected wealth plus a function of the goals. We introduce a simple scheme based on subjective priority levels to enforce and then estimate after solution the probability to attain the targets. The scheme is applied to an optimization problem with stochastic goals, random asset returns and liability costs, random income and consumption costs, as well as random retirement income upon retirement. Computational evidence is presented for the case of 2 or 5 goals and we show that solution times of a set of large scale linear programs do not depend on the number of goals, nor do goal achievement probabilities depend on the adopted mean-CVaR trade-off.
- 25.11.2026 | 15:40 | Tomáš Mrkvička
University of South Bohemia in České Budějovice, Czech Republic
Title: TBD
....
- 09.12.2026 | 15:40 | Vera Djordjilovic
Ca' Foscari University, Italy
Title: TBD
....
- 06.01.2027 | 15:40 | TBD
Title: TBD
....
Advanced Statistical Seminar (previous semesters)
|
The archive (seminar speakers, talk titles, and the abstracts) of the Advanced statistical seminar (NMST611)
from previous semesters. |
|
| Winter term 2025/2026 | Summer term 2026 |
|---|---|
| Winter term 2024/2025 | Summer term 2025 |
| Winter term 2023/2024 | Summer term 2024 |
| Summer term 2023 | |
| Summer term 2022 | |
| Summer term 2021 | |
