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

Forskningsoutput: Working paper

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.

Detaljer

Författare
Enheter & grupper
Forskningsområden

Ämnesklassifikation (UKÄ) – OBLIGATORISK

  • Sannolikhetsteori och statistik
Originalspråkengelska
Antal sidor13
StatusUnpublished - 2019
PublikationskategoriForskning

Related projects

Umberto Picchini, Julie Forman, Kresten Lindorff-Larsen & Samuel Wiqvist

2015/01/01 → …

Projekt: ForskningInternationellt samarbete, Tvärvetenskaplig forskning

Visa alla (1)