Dienstag, 18.04.2023, 12-13 Uhr in W9-109

Prof. Dr. Dietmar Bauer
Fakultät für Wirtschaftswissenschaften der Universität Bielefeld

Lag Length Selection Using Information Criteria for Integrated Processes

Information Criteria like AIC or BIC are often used for the specification of the lag length in autoregressive models. The properties of the corresponding estimators are well known in the case of stationary processes. The same procedures, however, are also used in the case where the processes are integrated of first or second order and hence show random walk like properties. For this case currently there is no correct result published clarifying the asymptotic properties of the selected lag lengths. This talk shows that the properties for the stationary case also carry over to the integrated case. Thus in the inverstible ARMA case the lag length of an autoregression selected using BIC increases proportional to log T asymptotically where the proportionality constant depends on the location of the zero closest to the unit circle.

Dienstag, 02.05.2023, 12-13 Uhr in W9-109

Carlina Feldmann, M.Sc.
Fakultät für Wirtschaftswissenschaften der Universität Bielefeld

Nonparametric inference in periodically inhomogeneous hidden Markov models

Within the class of hidden Markov models (HMMs), which is a popular tool for modelling time series driven by underlying states, periodic variation in the state-switching dynamics is routinely modelled using trigonometric functions. This parametric modelling can be too inflexible to capture complex periodic patterns, e.g. featuring multiple activity peaks per day.  In this talk, an alternative approach using cyclic penalised splines to model periodic variation within HMMs is presented. The challenge of estimating the corresponding complex models is substantially reduced by the expectation-maximisation algorithm, which allows to make use of the existing machinery (and software) for nonparametric regression. The practicality and potential usefulness of this approach is demonstrated in a real-data application modelling the activity of fruitflies.

Dienstag, 16.05.2023, 12-13 Uhr in W9-109

Dr. Jinglan Zheng
Fakultät für Physik der Universität Bielefeld

Statistics Development in Astronomy and Cosmology: Opportunities and Challenges

In recent years, the development of radio telescopes such as the Low Frequency Array (LOFAR) and the Square Kilometre Array (SKA) has generated an enormous amount of data, which presents significant challenges in data storage and processing. Statistical methods have become increasingly important in the analysis of this big data in astronomy and cosmology. In this talk, I will discuss the scientific problems in astronomy and cosmology that can be addressed using statistical methods, including parametric and non-parametric methods, Bayesian inference, and machine learning. Additionally, I will highlight the opportunities that arise from the Big Bang to Big Data (B3D) cluster in NRW. Universität Bielefeld is one of the coordination institutions and focuses on building a graduate school that connects astronomy to data science. With interdisciplinary approaches, there is tremendous potential for collaboration and cutting-edge discoveries in the fields of astronomy and data science.

Dienstag, 30.05.2023, 12-13 Uhr in W9-109

Prof. Dr. Göran Kauermann
Institut für Statistik der Ludwig-Maximilians- Universität München

Categorizing the World into Local Climate Zones – Towards Quantifying Labeling Uncertainty for Machine Learning Models

Image classification is often prone to labelling uncertainty. To generate suitable training data, images are labelled according to evaluations of human experts. This can result in ambiguities, which will affect subsequent models. In this work, we aim to model the labelling uncertainty in the context of remote sensing and the classification of satellite images. We construct a multinomial mixture model given the evaluations of multiple experts. This is based on the assumption that there is no ambiguity of the image class, but apparently in the experts' opinion about it. The model parameters can be estimated by a stochastic EM algorithm. Analyzing the estimates gives insights into sources of label uncertainty. Here, we focus on the general class ambiguity, the heterogeneity of experts, and the origin city of the images. The results are relevant for all machine learning applications where image classification is pursued and labelling is subject to humans.

Dienstag, 13.06.2023, 12-13 Uhr in W9-109

Karsten Reichold, M.Sc.
Fakultät Statistik der Technischen Universität Dortmund

Smooth Transition Cointegrating Regressions: Modified Nonlinear Least Squares Estimation and Inference

We develop fully modified and dynamic nonlinear least squares estimators for smooth transition cointegrating regressions that include deterministic and integrated variables as regressors and an integrated variable or time as transition variable. The stationary errors are allowed to be serially correlated and the regressors as well as the transition variable are allowed to be endogenous. Both estimators are shown to have the same zero-mean Gaussian mixture limiting distribution that allows for asymptotic standard inference. The theoretical analysis is complemented by a simulation study showing that the performance advantages of the modified estimators over nonlinear least squares are comparable to the performance advantages observed in linear cointegrating regressions. Finally, we use the developed methodology to investigate potential nonlinearities of long-run US money demand.

Dienstag, 27.06.2023, 12-13 Uhr in W9-109

Prof. Dr. Simon Kühne
Fakultät für Soziologie der Universität Bielefeld

Gender Stereotypes on Instagram: A Case Study Using Pose Estimation

In the contemporary digital era, social media has emerged as a significant source of behavioral data, reflecting real-world human actions in a virtual context. Despite extensive research utilizing textual social media data to understand individual attitudes, opinions, and traits, some characteristics such as gender role attitudes remain elusive, implicitly exhibited and challenging to measure. As an alternative to textual data, images, especially those shared on platforms like Instagram, may provide insightful perspectives to better understand these complex phenomena. Historically, gendered self-portrayal in photographs has reflected societal expectations and stereotypes of masculinity and femininity, an observation supported by established sociological research (Goffman 1979, Kang 1997, Götz & Becker 2019). Our study aims to harness social media images, particularly self-portraits, as a tool to examine gender roles, presentation, and stereotypes. Using Instagram images collected in 2018, we propose a novel approach to quantify gender portrayal, advancing beyond qualitative and manual techniques. By utilizing an automatic body pose detection algorithm, we identify the 2-dimensional skeletons of individuals within images. We then cluster these skeletons based on the similarity of their body pose. As a result, we obtain several clusters reflecting stereotypically gendered poses, some of which correspond with categories initially proposed by Goffman (1979) and Kang (1997). Moreover, we discover novel poses peculiar to social media and contemporary technologies, such as smartphones and 'selfies'. Our study offers a novel, automated method for quantitatively assessing gender stereotypes in self-portraits by examining body poses. Furthermore, our findings enhance our understanding of how online and social media platforms contribute to the propagation of gender stereotypes.

Dienstag, 11.07.2023, 12-13 Uhr in W9-109

Aktuelle Forschungsthemen im ZeSt