This course will take place as an online event. The link to the online platform will be provided to the accepted registrants only.
When observing data, the key question is: What can I learn from the observation? Bayesian inference treats all parameters of the model as random variables. The main task is to update their distribution as new data is observed. Hence, quantifying uncertainty of the parameter estimation is always part of the task. In this course we will introduce the basic theoretical concepts of Bayesian Statistics and Bayesian inference. We discuss the computational techniques and their implementations, different types of models as well as model selection procedures. We will exercise on the existing datasets use the PyMC3 framework for practicals.
The main topics are:
- Bayes theorem
- Prior and Posterior distributions
- Computational challenges and techniques: MCMC, variational approaches
- Models: Mixture Models, Bayesian Neural Networks, Variational Autoencoder, Normalizing Flows
- PyMC3 framework for Bayesian computation
- Running Bayesian models on a Supercomputer
Prerequisites:
Participants should be familiar with general statistical concepts, such as distributions, samples. Furthermore, familiarity with fundamental Machine Learning concepts such as regression, classification and training is helpful.
Language:
This course is given in English.
Date:
24-28 March 2025, 09:00-13:00 each day
Further information:
please visit the JSC webpage at https://go.fzj.de/2025-bayesian-sl
Instructor:
Dr. Alina Bazarova