Title:	ABC Inference for Multidimensional Diffusions Using Structure-preserving Integrators
DNr:	NAISS 2024/22-1213
Project Type:	NAISS Small Compute
Principal Investigator:	Petar Jovanovski <petarj@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2024-09-18 – 2025-10-01
Classification:	10106
Keywords:

Abstract

The goal of this project is to develop novel splitting methods for stochastic differential equations (SDE), tailored for use in a simulation-based inference framework to approximate posterior distributions. We additionally employ an invariant neural network, previously developed for Markov processes, to learn low-dimensional representations of SDE solutions. The neural network is incrementally retrained by exploiting an multi-round sampler, which provides new training data at each round