Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Partially Exchangeable Networks and Architectures for Learning Summary Statistics in Approximate Bayesian Computation
(Submitted on 29 Jan 2019 (v1), last revised 17 May 2019 (this version, v2))
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.
Submission history
From: Samuel Wiqvist [view email][v1] Tue, 29 Jan 2019 11:31:31 GMT (299kb,D)
[v2] Fri, 17 May 2019 14:19:59 GMT (572kb,D)
Link back to: arXiv, form interface, contact.