Learning Optimal Cyclic Causal Graphs from Interventional Data

Kari Rantanen, Antti Hyttinen, Matti Järvisalo
Proceedings of the 10th International Conference on Probabilistic Graphical Models, PMLR 138:365-376, 2020.

Abstract

We consider causal discovery in a very general setting involving non-linearities, cycles and several experimental datasets in which only a subset of variables are recorded. Recent approaches combining constraint-based causal discovery, weighted independence constraints and exact optimization have shown improved accuracy. However, they have mainly focused on the d-separation criterion, which is theoretically correct only under strong assumptions such as linearity or acyclicity. The more recently introduced sigma-separation criterion for statistical independence enables constraint-based causal discovery for non-linear relations over cyclic structures. In this work we make several contributions in this setting. (i) We generalize bcause, a recent exact branch-and-bound causal discovery approach, to this setting, integrating support for the sigma-separation criterion and several interventional datasets. (ii) We empirically analyze different schemes for weighting independence constraints in terms of accuracy and runtimes of bcause. (iii) We provide improvements to a previous declarative answer set programming (ASP) based approach for causal discovery employing the sigma-separation criterion, and empirically evaluate bcause and the refined ASP-approach.

Cite this Paper


BibTeX
@InProceedings{pmlr-v138-rantanen20a, title = {Learning Optimal Cyclic Causal Graphs from Interventional Data}, author = {Rantanen, Kari and Hyttinen, Antti and J\"arvisalo, Matti}, booktitle = {Proceedings of the 10th International Conference on Probabilistic Graphical Models}, pages = {365--376}, year = {2020}, editor = {Manfred Jaeger and Thomas Dyhre Nielsen}, volume = {138}, series = {Proceedings of Machine Learning Research}, month = {23--25 Sep}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v138/rantanen20a/rantanen20a.pdf}, url = { http://proceedings.mlr.press/v138/rantanen20a.html }, abstract = {We consider causal discovery in a very general setting involving non-linearities, cycles and several experimental datasets in which only a subset of variables are recorded. Recent approaches combining constraint-based causal discovery, weighted independence constraints and exact optimization have shown improved accuracy. However, they have mainly focused on the d-separation criterion, which is theoretically correct only under strong assumptions such as linearity or acyclicity. The more recently introduced sigma-separation criterion for statistical independence enables constraint-based causal discovery for non-linear relations over cyclic structures. In this work we make several contributions in this setting. (i) We generalize bcause, a recent exact branch-and-bound causal discovery approach, to this setting, integrating support for the sigma-separation criterion and several interventional datasets. (ii) We empirically analyze different schemes for weighting independence constraints in terms of accuracy and runtimes of bcause. (iii) We provide improvements to a previous declarative answer set programming (ASP) based approach for causal discovery employing the sigma-separation criterion, and empirically evaluate bcause and the refined ASP-approach.} }
Endnote
%0 Conference Paper %T Learning Optimal Cyclic Causal Graphs from Interventional Data %A Kari Rantanen %A Antti Hyttinen %A Matti Järvisalo %B Proceedings of the 10th International Conference on Probabilistic Graphical Models %C Proceedings of Machine Learning Research %D 2020 %E Manfred Jaeger %E Thomas Dyhre Nielsen %F pmlr-v138-rantanen20a %I PMLR %P 365--376 %U http://proceedings.mlr.press/v138/rantanen20a.html %V 138 %X We consider causal discovery in a very general setting involving non-linearities, cycles and several experimental datasets in which only a subset of variables are recorded. Recent approaches combining constraint-based causal discovery, weighted independence constraints and exact optimization have shown improved accuracy. However, they have mainly focused on the d-separation criterion, which is theoretically correct only under strong assumptions such as linearity or acyclicity. The more recently introduced sigma-separation criterion for statistical independence enables constraint-based causal discovery for non-linear relations over cyclic structures. In this work we make several contributions in this setting. (i) We generalize bcause, a recent exact branch-and-bound causal discovery approach, to this setting, integrating support for the sigma-separation criterion and several interventional datasets. (ii) We empirically analyze different schemes for weighting independence constraints in terms of accuracy and runtimes of bcause. (iii) We provide improvements to a previous declarative answer set programming (ASP) based approach for causal discovery employing the sigma-separation criterion, and empirically evaluate bcause and the refined ASP-approach.
APA
Rantanen, K., Hyttinen, A. & Järvisalo, M.. (2020). Learning Optimal Cyclic Causal Graphs from Interventional Data. Proceedings of the 10th International Conference on Probabilistic Graphical Models, in Proceedings of Machine Learning Research 138:365-376 Available from http://proceedings.mlr.press/v138/rantanen20a.html .

Related Material