Regret-based Federated Causal Discovery

Osman Mian, David Kaltenpoth, Michael Kamp
Proceedings of The KDD'22 Workshop on Causal Discovery, PMLR 185:61-69, 2022.

Abstract

In critical applications, causal models are the prime choice for their trustworthiness and explainability. If data is inherently distributed and privacy-sensitive, federated learning allows for collaboratively training a joint model. Such approaches for federated causal discovery, however, require sending local causal models, revealing the local data structure. We propose privacy-preserving federated causal discovery by distributed min-max regret optimization. This technique requires clients to only send local regret values, instead of model parameters, ensuring the privacy of sensitive local data. Initial results show that our approach reliably discovers causal networks without ever looking at local data or local causal structures.

Cite this Paper

BibTeX
@InProceedings{pmlr-v185-mian22a, title = {Regret-based Federated Causal Discovery}, author = {Mian, Osman and Kaltenpoth, David and Kamp, Michael}, booktitle = {Proceedings of The KDD'22 Workshop on Causal Discovery}, pages = {61--69}, year = {2022}, editor = {Le, Thuc Duy and Liu, Lin and Kıcıman, Emre and Triantafyllou, Sofia and Liu, Huan}, volume = {185}, series = {Proceedings of Machine Learning Research}, month = {15 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v185/mian22a/mian22a.pdf}, url = {https://proceedings.mlr.press/v185/mian22a.html}, abstract = {In critical applications, causal models are the prime choice for their trustworthiness and explainability. If data is inherently distributed and privacy-sensitive, federated learning allows for collaboratively training a joint model. Such approaches for federated causal discovery, however, require sending local causal models, revealing the local data structure. We propose privacy-preserving federated causal discovery by distributed min-max regret optimization. This technique requires clients to only send local regret values, instead of model parameters, ensuring the privacy of sensitive local data. Initial results show that our approach reliably discovers causal networks without ever looking at local data or local causal structures.} }
Endnote
%0 Conference Paper %T Regret-based Federated Causal Discovery %A Osman Mian %A David Kaltenpoth %A Michael Kamp %B Proceedings of The KDD'22 Workshop on Causal Discovery %C Proceedings of Machine Learning Research %D 2022 %E Thuc Duy Le %E Lin Liu %E Emre Kıcıman %E Sofia Triantafyllou %E Huan Liu %F pmlr-v185-mian22a %I PMLR %P 61--69 %U https://proceedings.mlr.press/v185/mian22a.html %V 185 %X In critical applications, causal models are the prime choice for their trustworthiness and explainability. If data is inherently distributed and privacy-sensitive, federated learning allows for collaboratively training a joint model. Such approaches for federated causal discovery, however, require sending local causal models, revealing the local data structure. We propose privacy-preserving federated causal discovery by distributed min-max regret optimization. This technique requires clients to only send local regret values, instead of model parameters, ensuring the privacy of sensitive local data. Initial results show that our approach reliably discovers causal networks without ever looking at local data or local causal structures.
APA
Mian, O., Kaltenpoth, D. & Kamp, M.. (2022). Regret-based Federated Causal Discovery. Proceedings of The KDD'22 Workshop on Causal Discovery, in Proceedings of Machine Learning Research 185:61-69 Available from https://proceedings.mlr.press/v185/mian22a.html.

Related Material