Partially exchangeable networks and architectures for learning summary statistics in approximate Bayesian computation
Paper in proceeding, 2019

We present a novel family of deep neural architectures, named partially exchangeable networks (PENs) that leverage probabilistic symmetries. By design, PENs are invariant to block-switch transformations, which characterize the partial exchangeability properties of conditionally Markovian processes. Moreover, we show that any block-switch invariant function has a PEN-like representation. The DeepSets architecture is a special case of PEN and we can therefore also target fully exchangeable data. We employ PENs to learn summary statistics in approximate Bayesian computation (ABC). When comparing PENs to previous deep learning methods for learning summary statistics, our results are highly competitive, both considering time series and static models. Indeed, PENs provide more reliable posterior samples even when using less training data.

Author

Samuel Wiqvist

Lund University

Pierre Alexandre Mattei

IT University of Copenhagen

Umberto Picchini

University of Gothenburg

Chalmers, Mathematical Sciences, Applied Mathematics and Statistics

Jes Frellsen

IT University of Copenhagen

36th International Conference on Machine Learning, ICML 2019

Vol. 2019-June 11795-11804

36th International Conference on Machine Learning, ICML 2019
Long Beach, USA,

Subject Categories

Other Computer and Information Science

Bioinformatics (Computational Biology)

Probability Theory and Statistics

More information

Latest update

1/29/2021