SUPR
Semi-amortised Bayesian inference for hierarchical mixed-effects stochastic models
Dnr:

NAISS 2024/22-1324

Type:

NAISS Small Compute

Principal Investigator:

Henrik Häggström

Affiliation:

Chalmers tekniska högskola

Start Date:

2024-10-14

End Date:

2025-04-01

Primary Classification:

10106: Probability Theory and Statistics

Webpage:

Allocation

Abstract

The goal of this project is to develop computationally efficient Bayesian inference methods for mixed-effects stochastic models, in particular we consider models with time-dynamics, with a focus on stochastic differential equations (SDE). Existing inference methods for these models are computationally intensive, which proves to be a computational bottleneck when the size of the data set increases. We develop a simulation-based inference method training mixture models to provide surrogates of the intractable likelihood and posterior.