Solving stochastic inverse problems with stochastic BayesFlow
- Normalizing flows have gained increasing attention in the area of probabilistic modeling. For solving inverse problems, BayesFlow is a state-of-the-art Bayesian inference method based on normalizing flows. However, BayesFlow suffers from overfitting in many real-world scenarios. Therefore, we put forward stochastic BayesFlow, enhancing BayesFlow through stochastic normalizing flows. Apart from being less prone to overfitting, stochastic BayesFlow performs more robustly in parameter identification from noisy observations. Moreover, we develop a stochastic BayesFlow algorithm to solve stochastic inverse problems and validate it using the inverse uncertainty quantification of a simulated vehicle dynamics model.
| Author: | Yi ZhangORCiDGND, Lars MikelsonsORCiDGND |
|---|---|
| Frontdoor URL | https://opus.bibliothek.uni-augsburg.de/opus4/119929 |
| ISBN: | 978-1-6654-7633-1OPAC |
| Parent Title (English): | 2023 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM), 28-30 June 2023, Seattle, WA, USA |
| Publisher: | IEEE |
| Place of publication: | Piscataway, NJ |
| Type: | Conference Proceeding |
| Language: | English |
| Year of first Publication: | 2023 |
| Contributing Corporation: | Seattle, WA, USA |
| Release Date: | 2025/03/11 |
| First Page: | 966 |
| Last Page: | 972 |
| DOI: | https://doi.org/10.1109/AIM46323.2023.10196190 |
| Institutes: | Fakultät für Angewandte Informatik |
| Fakultät für Angewandte Informatik / Institut für Informatik | |
| Fakultät für Angewandte Informatik / Institut für Informatik / Lehrstuhl für Ingenieurinformatik mit Schwerpunkt Mechatronik |
