Nov. December 2021 Jan. 2022
Modify filter

Filter off: No filter for categories


01.12.2021 12:30 Benito van der Zander (Universität zu Lübeck):
t.b.a.Online: attend

t.b.a.

06.12.2021 15:00 Dr. Merten Stender:
Recurrence-based nonlinear time series analysisvia ZOOM (Boltzmannstr. 3, 85748 Garching)

Evidence-based and data-driven techniques for nonlinear dynamical systems have matured in past decades. Following the era of chaotic dynamics in the 1980’s, nonlinear time series analysis has been expanding the toolset of signal processing techniques, now allowing to compute Lyapunov exponents and dimensionality metrics from measured time series data. This talk addresses the last major contribution to nonlinear analysis of time series data, namely the recurrence quantification analysis. These techniques are particularly relevant for the analysis of transient and multi-scale dynamics, and bridge time-domain analysis with network-based perspectives. The talk covers various aspects of recurrence analysis, correspondence to theoretic proofs, and exemplary case studies from the field of complex mechanical vibrations.

06.12.2021 16:30 Guilherme Henrique de Paula Reis (TUM):
TBAOnline: attendB 252 (Theresienstraße 39, 80333 München)

TBA

08.12.2021 12:30 Niels Richard Hansen (University of Copenhagen):
Conditional independence testing based on partial copulasOnline: attendBC1 2.01.10 (Parkring 11, 85748 Garching)

The partial copula provides a method for describing the dependence between two real valued random variables X and Y conditional on a third random vector Z in terms of nonparametric residuals. These residuals are in practice computed via models of the conditional distributions X|Z and Y|Z. In this talk I will show how the nonparametric residuals can be combined to give a valid test of conditional independence provided that nonparametric estimators of the conditional distributions converge at a sufficient rate. The rates can be realized via estimators based on quantile regression. If time permits, I will show how the test can be generalized to conditional local independence (Granger noncausality) in a time dynamic framework.

13.12.2021 15:00 Dr. Tommaso Rosati:
The Allen-Cahn equation with generic initial datumvia ZOOM (Boltzmannstr. 3, 85748 Garching)

We will study the scaling limit of the Allen-Cahn equation with generic random initial conditions. We will prove that after a suitably long time the dynamics are well approximated by a certain class of Gaussian nodal sets which evolve under mean curvature flow. The proofs build on Wild expansions of solutions to the Allen-Cahn equation. The talk is based on a joint work with Martin Hairer and Khoa Le

15.12.2021 12:30 Dennis Leung (University of Melbourne):
ZAP: z-value adaptive procedures for false discovery rate control with side informationOnline: attend

In the last five years, adaptive multiple testing with covariates has gained much traction. It has been recognized that the side information provided by auxiliary covariates which are independent of the primary test statistics under the null can be used to devise more powerful testing procedures for controlling the false discovery rate (FDR). For example, in the differential expression analysis of RNA-sequencing data, the average read counts across samples provide useful side information alongside individual p-values, as the genetic markers with higher read counts should be more promising to display differential expression.

However, for two-sided hypotheses, the usual data processing step that transforms the primary test statistics, generally known as z-values, into p-values not only leads to a loss of information carried by the main statistics but can also undermine the ability of the covariates to assist with the FDR inference. Motivated by this and building upon recent theoretical advances, we develop ZAP, a z-value based covariate-adaptive methodology. It operates on the intact structural information encoded jointly by the z-values and covariates, to mimic an oracle testing procedure that is unattainable in practice; the power gain of ZAP can be substantial in comparison with p-value based methods, as demonstrated by our simulations and real data analyses.

16.12.2021 17:00 Simon Gottschalk (UniBw München):
Reinforcement Learning and Classical Optimal Control - Links and SynergiesGebäude 33, Raum 1401 (Werner-Heisenberg-Weg 39, 85577 Neubiberg)

These days, machine learning techniques enter nearly every research field. Sometimes, because of this very fast growth, we forget to look back to classical successful methods. These well-established methods may already provide answers to recent challenges.

In this talk, we address this issue in the context of optimal control tasks (OC). We present a comparison of the Deep Reinforcement Learning (DRL) framework, representing the machine learning approach, and the classical theory. It turns out, that under mild assumptions the DRL framework can be transformed to an optimization problem, which is similar to an optimal control problem. This provides the opportunity to discuss classical results like the necessary optimality conditions in the context of DRL and to deduce new numerical methods.

Furthermore, we illustrate these results and further universal properties of RL by considering various applications cases. For example, we actuate muscles of a biomechanical human arm in order to reach a certain point and we find controls for a model of a satellite in order to perform a docking maneuver. Further examples are the steering of a car or a robot arm.

Besides the ability of RL to solve many quite different problems, it is shown that there are also some limitations such as a costly training for simple subtasks or the presence of many local optima, which are far away from the global solution. Some of these can be handled by classical OC approaches. Thus, in the end of this talk, we give an outlook on how a hybrid method, which unites advantages from the RL as well as from the OC framework, could extract the best out of both worlds.

20.12.2021 16:30 Tom Kaufmann (Universität Bochum):
TBAOnline: attendB 252 (Theresienstraße 39, 80333 München)

TBA