Partially exchangeable networks and architectures for learning summary statistics in approximate Bayesian computation
(2019) 36th International Conference on Machine Learning, ICML 2019 In 36th International Conference on Machine Learning, ICML 2019 2019-June. p.11795-11804- 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... (More)
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.
(Less)
- author
- Wiqvist, Samuel LU ; Mattei, Pierre Alexandre ; Picchini, Umberto LU and Frellsen, Jes
- organization
- publishing date
- 2019
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- 36th International Conference on Machine Learning, ICML 2019
- series title
- 36th International Conference on Machine Learning, ICML 2019
- volume
- 2019-June
- pages
- 10 pages
- publisher
- International Machine Learning Society (IMLS)
- conference name
- 36th International Conference on Machine Learning, ICML 2019
- conference location
- Long Beach, United States
- conference dates
- 2019-06-09 - 2019-06-15
- external identifiers
-
- scopus:85078252544
- ISBN
- 9781510886988
- language
- English
- LU publication?
- yes
- id
- 3c47b80e-9d83-4333-af07-d317d6bae77d
- date added to LUP
- 2020-02-07 13:08:59
- date last changed
- 2022-04-18 20:21:04
@inproceedings{3c47b80e-9d83-4333-af07-d317d6bae77d, abstract = {{<p>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.</p>}}, author = {{Wiqvist, Samuel and Mattei, Pierre Alexandre and Picchini, Umberto and Frellsen, Jes}}, booktitle = {{36th International Conference on Machine Learning, ICML 2019}}, isbn = {{9781510886988}}, language = {{eng}}, pages = {{11795--11804}}, publisher = {{International Machine Learning Society (IMLS)}}, series = {{36th International Conference on Machine Learning, ICML 2019}}, title = {{Partially exchangeable networks and architectures for learning summary statistics in approximate Bayesian computation}}, volume = {{2019-June}}, year = {{2019}}, }