The Interuniversity Institute for Biostatistics and statistical Bioinformatics organizes for the 8th time the Summer School on Advanced Bayesian Methods, which is on novel Bayesian methods relevant to the applied statistician. Two courses will be organized in Leuven from 14 to 18 September 2026:
- Two-day course (14-15 September) on Bayesian Finite Mixture Analysis by Prof. em. Sylvia Frühwirth-Schnatter (Vienna University of Economics and Business, Austria)
- Three-day course (16-18 September) on Bayesian Smoothing and Distributional Regression by Prof Thomas Kneib and Johannes Brachem (Georg-August-Universität Göttingen, Germany)
The target audience of the summer school are statisticians and/or epidemiologists with a sound background in statistics, but also with a background in Bayesian methodology. In both courses, practical sessions are organized, so participants are asked to bring along their laptop with the appropriate software (to be announced) pre-installed. The contents of the course and a biosketch of the instructors can be found below.
Location
The course will take place in Park Inn by Radisson hotel, Belgium (next to the train station). Lunch is included in the registration fee but a stay in the Park Inn by Radisson hotel is not included. If you need hotel accommodation, you need to arrange it yourself. Accommodation options can be found on the Visit Leuven site.
The registration costs for the courses are:
| Two-day course | Early bird (Full rate) | Three-day course | Early bird (Full rate) | |
| I-Biostat member | € 50 (75) | I-Biostat member | € 50 (75) | |
| PhD student | € 200 (250) | PhD student | € 250 (300) | |
| Quetelet member | € 200 (250) | Quetelet member | € 250 (300) | |
| Academic | € 300 (400) | Academic | € 400 (500) | |
| ISBA member | € 300 (400) | ISBA member | € 400 (500) | |
| Research institute | € 300 (400) | Research institute | € 400 (500) | |
| Industry | € 600 (900) | Industry | € 900 (1200) |
Early bird registrations (registration before May 31, 2026 and payment before June 30, 2026) enjoy the reduced rate. In any case, registration should be before August 1, 2026 and payment before August 31, 2026 (at the full rate). Registration is now closed.
Cancellation policy
If cancelled before 31 July 2026, a full refund will be given.
If cancelled after 31 July 2026 but before 31 August 2026, a 50% refund will be given.
For cancellations after 31 August 2026, no refund will be given.
For questions about the registration costs, please contact Romy Gentens. For other questions, please contact Emmanuel Lesaffre.
Supported by:
 |
|
|
Course descriptions
Bayesian Finite Mixture Analysis
Sylvia Frühwirth-Schnatter (Vienna University of Economics and Business, Austria)
Since Karl Pearson’s seminal work in 1894, finite mixture models have attracted a lot of attention due to their versatility in analyzing statistical data. Finite mixture models allow a flexible and robust data analysis in situations where a single probability distribution is not adequate to describe the data. They allow to infer latent heterogeneity in possibly large data sets and have a wide range of applications. A particularly popular application of finite mixture analysis arises in model-based clustering. A Bayesian treatment of finite mixture models is attractive for several reasons, in particular when the number of components of the mixture distribution is unknown. It allows to obtain uncertainty quantification not only for the parameters of the component distributions, but also for the number of clusters in the data and the number of mixture components in the underlying population.
Starting with a basic introduction into finite mixture analysis, this course provides a comprehensive overview over the Bayesian approach toward finite mixture analysis. The course allows the students to gain and broaden their knowledge in three aspects of Bayesian finite mixture analysis:
- Applied finite mixture analysis. Throughout the course, applications to a wide range of data sets and component models are discussed. Starting with univariate and multivariate Gaussian mixtures as well as latent-class models, finite mixture analysis is expanded in several ways to achieve greater flexibility. These extensions range from mixtures of skew densities, mixtures-of-experts models, mixtures of finite mixtures to mixtures of factor analyzers.
- Computational techniques. The course covers computational aspects and discusses various Markov Monte Carlo simulation techniques for the various types of finite mixture models, both when the number of mixture components is known and unknown. Special emphasis is placed on post-processing procedures to resolve the label switching problem arising during posterior sampling.
- The impact of prior choices. The course provides a comprehensive understanding of prior choices in finite mixture analysis and their impact on posterior inference. A major focus is placed on the Dirichlet prior distribution on the mixture weight distribution and how it impacts the prior distribution on the partition of the data and the number of data clusters. These insights will allow the students not only to control the impact of prior choices, but also to understand the close connections between finite mixtures and Bayesian non-parametric mixtures such as Dirichlet process and Pitman-Yor process mixtures.
Bayesian Smoothing and Distributional Regression
Thomas Kneib and Johannes Brachem (Georg-August-Universität Göttingen, Germany)
Flexible forms of regression modelling such as generalized additive models, random effects models, or spatial smoothing have received enduring attention in the past decades. They inherit the generality in terms of response types from generalized linear models and additionally allow for nonlinear effects of continuous covariates, cluster-specific heterogeneity, and spatial heterogeneity, respectively. Recently, these models have been extended towards distributional regression modelling, where not only the conditional mean of a response variable, but also other aspects of the response distribution such as variability or skewness are related to explanatory variables. For the resulting class of additive distributional regression models, a Bayesian treatment is particularly attractive:
- For the different model components, assumptions on their smoothness can be conveniently implemented and controlled via suitable prior distributions.
- A computationally convenient divide-and-conquer strategy can be achieved with Markov chain Monte Carlo (MCMC) simulation techniques, benefitting both the development of models and their actual implementation in statistical software.
- MCMC simulations also facilitate uncertainty quantification without the need to resort to asymptotic considerations not only for the original model parameters but also for complex functionals derived from them.
In this course, we will introduce Bayesian smoothing and distributional regression in three main blocks:
- In the first block on Bayesian additive models, we will introduce Bayesian penalized splines as a convenient and versatile tool for representing complex, nonlinear shapes of covariate effects. We will also discuss associated ways of conducting Bayesian inference and the choice of hyperparameters and suitable hyperpriors associated with them.
- The second block on Bayesian structured additive regression will embed Bayesian additive models in a broader class of models that provides a unifying framework for various effect types including penalized splines, spatial effects and random effects as special cases. This considerably broadens the model class and the types of applications that can be considered.
- Finally, the third block on Bayesian structure additive distributional regression will add a distributional perspective to the models where not only the mean but also other distributional features can be related to explanatory variables based on structured additive regression predictors.
In all three blocks, we will discuss the methodological foundations, Bayesian inference based on MCMC simulations, the implementation in statistical software, and the practical application in terms of case studies. Computer labs supplement the lecture components such that the attendees gain first-hand experience in applying the models presented in the course.
Bio sketches of the course instructors
Prof. Sylvia Frühwirth-Schnatter
Sylvia Frühwirth-Schnatter is a recently retired Professor of Applied Statistics and Econometrics at the Vienna University of Economics and Business (Austria). Her research focuses on Bayesian modeling and Markov chain Monte Carlo inference for a broad range of statistical models, including finite mixtures analysis, state space modeling and sparse factor analysis. She is particularly interested in applications of Bayesian inference in economics, finance, and business. She was President of the International Society for Bayesian Analysis (ISBA) and was recently appointed Co-Editor of the Journal of Applied Econometrics. Her 2006 monograph on Finite Mixture and Markov Switching Models was awarded the Morris-DeGroot Price by ISBA. She is elected Member of the Austrian Academy of Sciences and ISBA Fellow . In 2024, she was awarded the Zellner Medal by ISBA.
Prof. Thomas Kneib and Johannes Brachem
Thomas Kneib is Professor for Statistics at Georg-August-Universität Göttingen, Germany. His research focuses on semiparametric regression models including distributional regression, random effects models and spatial statistics as well as statistical learning including regularized maximum likelihood inference, Bayesian inference and functional gradient descent boosting. He is deputy spokesperson of Göttingen’s Campus Institute Data Science and co-editor of the Journal of the Royal Statistical Society, Series C (Applied Statistics).
Johannes Brachem is a PhD student at the Chair of Statistics at Georg-August-Universität Göttingen, Germany. His work focuses on Bayesian transformation models, a new form of semiparametric distributional regression. He is part of the development team of Liesel, a Python probabilistic programming framework for Bayesian distributional regression (https://liesel-project.org/).