Combining formal methods and Bayesian approach for inferring discrete-state stochastic models from steady-state data

Investor logo

Warning

This publication doesn't include Faculty of Sports Studies. It includes Faculty of Informatics. Official publication website can be found on muni.cz.
Authors

KLEIN Julia PHUNG Huy HAJNAL Matej ŠAFRÁNEK David PETROV Tatjana

Year of publication 2023
Type Article in Periodical
Magazine / Source Plos one
MU Faculty or unit

Faculty of Informatics

Citation
Web https://doi.org/10.1371/journal.pone.0291151
Doi http://dx.doi.org/10.1371/journal.pone.0291151
Keywords Bayes Theorem; Markov Chains
Description Stochastic population models are widely used to model phenomena in different areas such as cyber-physical systems, chemical kinetics, collective animal behaviour, and beyond. Quantitative analysis of stochastic population models easily becomes challenging due to the combinatorial number of possible states of the population. Moreover, while the modeller easily hypothesises the mechanistic aspects of the model, the quantitative parameters associated to these mechanistic transitions are difficult or impossible to measure directly. In this paper, we investigate how formal verification methods can aid parameter inference for population discrete-time Markov chains in a scenario where only a limited sample of population-level data measurements-sample distributions among terminal states-are available. We first discuss the parameter identifiability and uncertainty quantification in this setup, as well as how the existing techniques of formal parameter synthesis and Bayesian inference apply. Then, we propose and implement four different methods, three of which incorporate formal parameter synthesis as a pre-computation step. We empirically evaluate the performance of the proposed methods over four representative case studies. We find that our proposed methods incorporating formal parameter synthesis as a pre-computation step allow us to significantly enhance the accuracy, precision, and scalability of inference. Specifically, in the case of unidentifiable parameters, we accurately capture the subspace of parameters which is data-compliant at a desired confidence level.
Related projects:

You are running an old browser version. We recommend updating your browser to its latest version.

More info