Robust Causal Domain Adaptation in a Simple Diagnostic Setting

Thijs van Ommen
Proceedings of the Eleventh International Symposium on Imprecise Probabilities: Theories and Applications, PMLR 103:424-429, 2019.

Abstract

Causal domain adaptation approaches aim to find statistical relations in a source domain, that will still hold in a target domain, using the assumption that a common causal graph underlies both domains. For many such problems, the available information is insufficient to uniquely identify the target domain distribution, and we find a set of distributions instead. We propose to use a worst-case approach, picking an action that performs well against all distributions in this set. In this paper, we study a specific diagnostic instance of this problem, and find a sufficient and necessary condition that characterizes the worst-case distribution in the target domain. We find that the Brier and logarithmic scores lead to different distributions, and consequently to different recommendations for the decision maker.

Cite this Paper

BibTeX
@InProceedings{pmlr-v103-van-ommen19a, title = {Robust Causal Domain Adaptation in a Simple Diagnostic Setting}, author = {{van Ommen}, Thijs}, booktitle = {Proceedings of the Eleventh International Symposium on Imprecise Probabilities: Theories and Applications}, pages = {424--429}, year = {2019}, editor = {De Bock, Jasper and de Campos, Cassio P. and de Cooman, Gert and Quaeghebeur, Erik and Wheeler, Gregory}, volume = {103}, series = {Proceedings of Machine Learning Research}, month = {03--06 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v103/van-ommen19a/van-ommen19a.pdf}, url = {https://proceedings.mlr.press/v103/van-ommen19a.html}, abstract = {Causal domain adaptation approaches aim to find statistical relations in a source domain, that will still hold in a target domain, using the assumption that a common causal graph underlies both domains. For many such problems, the available information is insufficient to uniquely identify the target domain distribution, and we find a set of distributions instead. We propose to use a worst-case approach, picking an action that performs well against all distributions in this set. In this paper, we study a specific diagnostic instance of this problem, and find a sufficient and necessary condition that characterizes the worst-case distribution in the target domain. We find that the Brier and logarithmic scores lead to different distributions, and consequently to different recommendations for the decision maker.} }
Endnote
%0 Conference Paper %T Robust Causal Domain Adaptation in a Simple Diagnostic Setting %A Thijs van Ommen %B Proceedings of the Eleventh International Symposium on Imprecise Probabilities: Theories and Applications %C Proceedings of Machine Learning Research %D 2019 %E Jasper De Bock %E Cassio P. de Campos %E Gert de Cooman %E Erik Quaeghebeur %E Gregory Wheeler %F pmlr-v103-van-ommen19a %I PMLR %P 424--429 %U https://proceedings.mlr.press/v103/van-ommen19a.html %V 103 %X Causal domain adaptation approaches aim to find statistical relations in a source domain, that will still hold in a target domain, using the assumption that a common causal graph underlies both domains. For many such problems, the available information is insufficient to uniquely identify the target domain distribution, and we find a set of distributions instead. We propose to use a worst-case approach, picking an action that performs well against all distributions in this set. In this paper, we study a specific diagnostic instance of this problem, and find a sufficient and necessary condition that characterizes the worst-case distribution in the target domain. We find that the Brier and logarithmic scores lead to different distributions, and consequently to different recommendations for the decision maker.
APA
van Ommen, T.. (2019). Robust Causal Domain Adaptation in a Simple Diagnostic Setting. Proceedings of the Eleventh International Symposium on Imprecise Probabilities: Theories and Applications, in Proceedings of Machine Learning Research 103:424-429 Available from https://proceedings.mlr.press/v103/van-ommen19a.html.

Related Material