Advancing Variational Inference: Semi-Implicit Models, Adaptive Proposals, and Functional Stein Gradients

MCML Authors

Tobias Pielok

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This dissertation advances variational inference for Bayesian machine learning by developing more expressive and stable approximation methods. It introduces function-space optimization techniques, improved semi-implicit variational inference with importance sampling, and kernel-based gradient estimators, enabling more accurate posterior and predictive inference in complex models such as Bayesian neural networks. (Shortened.)

phdthesis Pie26

---

Dissertation

LMU München. Mar. 2026

---

Authors

T. Pielok

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability