Partially Exchangeable Networks and Architectures for Learning Summary Statistics in Approximate Bayesian Computation
Samuel Wiqvist · Pierre-Alexandre Mattei · Umberto Picchini · Jes Frellsen

Thu Jun 13th 05:00 -- 05:05 PM @ Grand Ballroom

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 Information

Samuel Wiqvist (Lund University)
Pierre-Alexandre Mattei (IT University Copenhagen)
Umberto Picchini (Department of Mathematical Sciences, Chalmers University of Technology and the University of Gothenburg)
Jes Frellsen (IT University of Copenhagen)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors