Bayesian methods in Machine Learning

Bayesian methods have recently regained a significant amount of attention in the machine community due to the development of scalable approximate Bayesian inference techniques. There are several advantages of using Bayesian approach: Parameter and prediction uncertainty become easily available, facilitating rigid statistical analysis. Furthermore, prior knowledge can be incorporated.

Image may contain: Triangle, Slope, Parallel, Font, Circle.

About the project

In this project we develop new models, methods and algorithms for Bayesian machine learning, in particular related to deep neural networks and computational causal inference.

Sub-projects

  • Bayesian estimation of causal effects using directed graphical models. PhD project for Vera Kvisgaard.
  • A Bayesian model-averaging toolkit for causal inference with observational data under nonlinear structural equations: An application to the effect of ADHD treatment on school performance by Norwegian children. PhD project for Johan de Aguas.
  • Normalizing flows as variational inference approximations in latent binary Bayesian neural network models. Master project by Lars Skaaret-Lund
  • A fully probabilistic methodology for providing diverse, personalised recommendations from clicking data, using a Variational Bayesian approach for fast computation. Master student: Haakon Muggerud
  • Subsampling Strategies for Bayesian Variable Selection and Model Averaging in GLM and BGNLM. Master project by Jon Lachmann (Stockholm University)
  • Identification of non-linear Models with a Bayesian Model selection Tool. Master project by Elke Bruns (University of Vienna)

Financing

  • Aliaksandr Hubin is hired as a post doc through the "Akademia-avtale" with Equinor
  • Part of the activity is financed through BigInsight

Selected publications

​​​​​​

Tags: data science
Published Apr. 27, 2022 11:31 AM - Last modified June 6, 2022 12:19 PM