Differentiable Causal Discovery Under Latent Interventions

Gonçalo Rui Alves Faria, Andre Martins, Mario A. T. Figueiredo
Proceedings of the First Conference on Causal Learning and Reasoning, PMLR 177:253-274, 2022.

Abstract

Recent work has shown promising results in causal discovery by leveraging interventional data with gradient-based methods, even when the intervened variables are unknown. However, previous work assumes that the correspondence between samples and interventions is known, which is often unrealistic. We envision a scenario with an extensive dataset sampled from multiple intervention distributions and one observation distribution, but where we do not know which distribution originated each sample and how the intervention affected the system, \textit{i.e.}, interventions are entirely latent. We propose a method based on neural networks and variational inference that addresses this scenario by framing it as learning a shared causal graph among a infinite mixture (under a Dirichlet process prior) of intervention structural causal models . Experiments with synthetic and real data show that our approach and its semi-supervised variant are able to discover causal relations in this challenging scenario.

Cite this Paper

BibTeX
@InProceedings{pmlr-v177-faria22a, title = {Differentiable Causal Discovery Under Latent Interventions}, author = {Faria, Gon{\c{c}}alo Rui Alves and Martins, Andre and Figueiredo, Mario A. T.}, booktitle = {Proceedings of the First Conference on Causal Learning and Reasoning}, pages = {253--274}, year = {2022}, editor = {Schölkopf, Bernhard and Uhler, Caroline and Zhang, Kun}, volume = {177}, series = {Proceedings of Machine Learning Research}, month = {11--13 Apr}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v177/faria22a/faria22a.pdf}, url = {https://proceedings.mlr.press/v177/faria22a.html}, abstract = {Recent work has shown promising results in causal discovery by leveraging interventional data with gradient-based methods, even when the intervened variables are unknown. However, previous work assumes that the correspondence between samples and interventions is known, which is often unrealistic. We envision a scenario with an extensive dataset sampled from multiple intervention distributions and one observation distribution, but where we do not know which distribution originated each sample and how the intervention affected the system, \textit{i.e.}, interventions are entirely latent. We propose a method based on neural networks and variational inference that addresses this scenario by framing it as learning a shared causal graph among a infinite mixture (under a Dirichlet process prior) of intervention structural causal models . Experiments with synthetic and real data show that our approach and its semi-supervised variant are able to discover causal relations in this challenging scenario. } }
Endnote
%0 Conference Paper %T Differentiable Causal Discovery Under Latent Interventions %A Gonçalo Rui Alves Faria %A Andre Martins %A Mario A. T. Figueiredo %B Proceedings of the First Conference on Causal Learning and Reasoning %C Proceedings of Machine Learning Research %D 2022 %E Bernhard Schölkopf %E Caroline Uhler %E Kun Zhang %F pmlr-v177-faria22a %I PMLR %P 253--274 %U https://proceedings.mlr.press/v177/faria22a.html %V 177 %X Recent work has shown promising results in causal discovery by leveraging interventional data with gradient-based methods, even when the intervened variables are unknown. However, previous work assumes that the correspondence between samples and interventions is known, which is often unrealistic. We envision a scenario with an extensive dataset sampled from multiple intervention distributions and one observation distribution, but where we do not know which distribution originated each sample and how the intervention affected the system, \textit{i.e.}, interventions are entirely latent. We propose a method based on neural networks and variational inference that addresses this scenario by framing it as learning a shared causal graph among a infinite mixture (under a Dirichlet process prior) of intervention structural causal models . Experiments with synthetic and real data show that our approach and its semi-supervised variant are able to discover causal relations in this challenging scenario.
APA
Faria, G.R.A., Martins, A. & Figueiredo, M.A.T.. (2022). Differentiable Causal Discovery Under Latent Interventions. Proceedings of the First Conference on Causal Learning and Reasoning, in Proceedings of Machine Learning Research 177:253-274 Available from https://proceedings.mlr.press/v177/faria22a.html.

Related Material