Partially Exchangeable Networks and Architectures for Learning Summary Statistics in Approximate Bayesian Computation

Abstract

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.

Forthcoming on the Proceedings of ICML 2019. New comparisons with several different networks. We now use the Wasserstein distance to produce comparisons. Code available on GitHub. 16 pages, 5 figures, 21 tables

Cited by in corpus (8)

• not-MIWAE: Deep Generative Modelling with Missing not at Random Data
• Neural Approximate Sufficient Statistics for Implicit Models
• ABCpy: A High-Performance Computing Perspective to Approximate Bayesian Computation
• Cosmological Parameter Estimation and Inference using Deep Summaries
• Sequential Neural Posterior and Likelihood Approximation
• Stratified sampling and bootstrapping for approximate Bayesian computation
• Towards Data-Conditional Simulation for ABC Inference in Stochastic Differential Equations
• Learning summary features of time series for likelihood free inference