VVZ API is not affiliated with ETH Zurich. Data might be outdated or incorrect. Please view the official ETHZ Vorlesungsverzeichnis for binding information.
Bayesian Model Selection in Theory and Practice
Last Updated: 2026-02-05 16:38:08
Abstract
An inverse problem in geoscience can be viewed as a process of integrating data, physical law and geological a priori information. The unavoidable uncertainties in all these types of information call for a probabilistic treatment, but how do we choose the basic setup of the problem, such as the number of parameters, data uncertainty and a priori information in the parameter space?
Objective
The aim of the course is to give the student insight into the difficulties and opportunities of advanced probabilistic inverse problems. The course will enable the successful student to - design a geophysical experiment aimed at acquiring maximum information about model parameters of interest - build reliable probability models for data uncertainties - analyze forward modeling uncertainties and evaluate their impact on solutions to the inverse problem - construct probabilistic a priori models - use samples form the posterior probability distribution to analyze solutions to the inverse problem
Content
The course consists of 9 parts: 1. Introduction and examples 2. Parameterization of physical systems. Information and probabilistic models. 3. Data uncertainties. What do they mean, and how do we obtain them? 4. Quantification of forward modeling errors. Are such errors important, and how can they be estimated? 5. Construction of geological a priori models 6. Classical Bayesian Inversion and the method of Tarantola and Valette 7. Model selection 1: Occam's Razor. Trans-Dimensional Inversion 8. Model selection 2: Hierarchical Bayes - Bayesian Inversion without priors in model and data space? 9. Optimization of data acquisition - Extraction of information from samples of the posterior probability distribution
General Information
- Language
- English
- Levels
- MSC
- Frequency
- Yearly recurring
Examination
- Type
- ungraded semester performance
Course Components
| Type | Title | Time & Place | Hours |
|---|---|---|---|
| lecture |
Bayesian Model Selection in Theory and Practice
Taught by visiting professor Klaus Mosegaard
The course starts on 29 February 2024.
|
|
2 h weekly |
Offered In
-
-
Electives (Courses can be chosen from the complete offerings of the ETH Zurich and University of Zurich (according to prior agreement with the MSc Committee).)
-