We especially congratulate Johannes Brachem from the Chair of Statistics at the University of Göttingen for the award for the "Best Student Talk"!
Quentin Seiferts presented his work on flexible regression using neural networks. The strengths of (semi-)parametric regression and neural networks were combined and emphasised in three different applications.
Alexandra Daub showed her adaptive step lengths in component-wise gradient boosting in a three-parameter distribution. As an application, the number of antenatal care visits in Nigeria has been modelled.
Lars Kniepers contribution Spatially aware gradient boosting towards sparser models focusses on the fair integration of spatial effects in gradient boosting. His approach has been illustrated in an application for predicting coffee harvests.
Poster contributions of
Tobias Hepp: Sparse spatial pattern selection via component-wise gradient boosting
Marisa Lange: Distributional Regression for Lungfunction of Cystic Fibrosis Patients with a Special Focus on Spatial and Random Effects
Sophie Potts: Joint Models for rare events
were represented.
The short articles of the contributions can be found here.
- Quentin Seiferts, Elisabeth Bergherr, Benjamin Säfken, Tobias Hepp:
Flexible Regression in Neural Networks
- Tobias Hepp, Anna von Plessen, Nadia Müller-Voggel, Elisabeth Bergherr:
Sparse spatial pattern selection via component-wise gradient boosting
- Lars Knieper, Elisabeth Bergherr:
Re-thinking spatial components in gradient boosting
- Alexandra Daub, Lars Knieper, Elisabeth Bergherr:
Balanced boosting for GAMLSS using adaptive step lengths with an application to antenatal care visits data in West African countries
- Colin Griesbach , Elisabeth Bergherr:
Boosting for Mixed Distributional Regression
- Sophie Potts, Elisabeth Bergherr:
Joint models for rare events
- John F. Brüne, Sophie Potts, Elisabeth Bergherr:
Silent Partys: A Cluster Analysis of Voting Behavior in the European Parliament
Using the API of abgeordnetenwatch.de, data was extracted, processed and analysed descriptively and multivariate using statistical methods. For example, groups of parliamentarians were identified based on their voting behaviour. The results are visualised in a user-friendly interactive app and help to communicate scientific findings from data to a broad public.
The poster and its presentation convinced the participants, who voted and awarded John the ‘Best Poster Prize’. The award-winning poster can be found in the hallway and you can try out the underlying app here. Congratulations, John!
Alexandra Daub presented an approach on how gradient boosting methods for GAMLSS models can be improved by adaptive step lengths. Her method, which is based on the idea of establishing a balance between the individual base learners, shows promising results in both simulated and real data sets. A preprint was published earlier this year.
Anna von Plessen presented her work, which deals with magnetoencephalography (MEG) data of brain waves. She is developing a concept to improve the processing of this high-dimensional data using methods such as gradient boosting and functional regression. As part of her presentation, she also showed her R Shiny application for visualising MEG data.
Colin Griesbach gave a presentation on model-based gradient boosting methods for GAMLSS models. In this context, he developed a method to improve the estimation of random effects. In his lecture, he presented the results of this method using data from cystic fibrosis patients.
Marisa Lange also works with data from cystic fibrosis patients. In her presentation, she gave a detailed insight into the structure and characteristics of the data and presented modelling approaches using gradient boosting. A particular focus of her work is the processing of spatial information in the data.
Dr. Colin Griesbach opened the conference with his presentation on model-based gradient boosting methods for GAMLSS models applied to data from cystic fibrosis patients. The conference article has been published in the conference proceedings.
Lars Knieper gave a presentation on the phenomenon of spatial confounding, which can be resolved in model-based gradient boosting using the spatial+ approach. He illustrated his approach using AirBnB data. The corresponding short article has been published by Springer.
Quentin Seiferts paper entitled ‘Function-on-scalar regression via first-order gradient-based optimization’ combines classical statistics with elements from neural networks and allows to analyse a large amount of parking data during the COVID-19 pandemic and identify patterns in shopping behaviour.
Dr. Joaquin Cavieres also presented part of his research comparing approximated Gaussian random fields with different parameterization approaches. The article has also been published.
Sophie Potts presented her work on the application of joint models for longitudinal and time-to-event data in the social sciences as a poster. The method, which originated in biostatistics, was applied to an example from family sociology. Sophie's novel application and accessible presentation was honoured with the ‘Best Student Poster Award’. The award-winning poster can now be admired in the corridor of our offices.
