Knowledge transfer for learning subject-specific causal models

Verónica Rodrı́guez-López, Luis Enrique Sucar
Proceedings of The 11th International Conference on Probabilistic Graphical Models, PMLR 186:385-396, 2022.

Abstract

Subject-specific causal models are appropriate for domains such as biology, medicine, and neuroscience, where the causal relations vary across the individuals of a population. However, its learning could be challenging, particularly under limited data sets. Although some works have addressed this issue, they are restricted to discovering up to Markov equivalence classes. In this work, we propose a method for the causal relations identification of subject-specific models. We hypothesized that transferring related data sets and locally performing interventions improves the causal direction identification of relations. The experimental results on true and imperfect Markov equivalence classes of synthetic causal Bayesian networks show that our method performing interventions over several subsets of the candidate parents and using related data according to their differences with targets recovers a higher number of correct oriented edges.

Cite this Paper

BibTeX
@InProceedings{pmlr-v186-rodri-guez-lopez22a, title = {Knowledge transfer for learning subject-specific causal models}, author = {Rodr\'{\i}guez-L\'{o}pez, Ver\'{o}nica and Sucar, Luis Enrique}, booktitle = {Proceedings of The 11th International Conference on Probabilistic Graphical Models}, pages = {385--396}, year = {2022}, editor = {Salmerón, Antonio and Rumı́, Rafael}, volume = {186}, series = {Proceedings of Machine Learning Research}, month = {05--07 Oct}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v186/rodri-guez-lopez22a/rodri-guez-lopez22a.pdf}, url = {https://proceedings.mlr.press/v186/rodri-guez-lopez22a.html}, abstract = {Subject-specific causal models are appropriate for domains such as biology, medicine, and neuroscience, where the causal relations vary across the individuals of a population. However, its learning could be challenging, particularly under limited data sets. Although some works have addressed this issue, they are restricted to discovering up to Markov equivalence classes. In this work, we propose a method for the causal relations identification of subject-specific models. We hypothesized that transferring related data sets and locally performing interventions improves the causal direction identification of relations. The experimental results on true and imperfect Markov equivalence classes of synthetic causal Bayesian networks show that our method performing interventions over several subsets of the candidate parents and using related data according to their differences with targets recovers a higher number of correct oriented edges.} }
Endnote
%0 Conference Paper %T Knowledge transfer for learning subject-specific causal models %A Verónica Rodrı́guez-López %A Luis Enrique Sucar %B Proceedings of The 11th International Conference on Probabilistic Graphical Models %C Proceedings of Machine Learning Research %D 2022 %E Antonio Salmerón %E Rafael Rumı́ %F pmlr-v186-rodri-guez-lopez22a %I PMLR %P 385--396 %U https://proceedings.mlr.press/v186/rodri-guez-lopez22a.html %V 186 %X Subject-specific causal models are appropriate for domains such as biology, medicine, and neuroscience, where the causal relations vary across the individuals of a population. However, its learning could be challenging, particularly under limited data sets. Although some works have addressed this issue, they are restricted to discovering up to Markov equivalence classes. In this work, we propose a method for the causal relations identification of subject-specific models. We hypothesized that transferring related data sets and locally performing interventions improves the causal direction identification of relations. The experimental results on true and imperfect Markov equivalence classes of synthetic causal Bayesian networks show that our method performing interventions over several subsets of the candidate parents and using related data according to their differences with targets recovers a higher number of correct oriented edges.
APA
Rodrı́guez-López, V. & Sucar, L.E.. (2022). Knowledge transfer for learning subject-specific causal models. Proceedings of The 11th International Conference on Probabilistic Graphical Models, in Proceedings of Machine Learning Research 186:385-396 Available from https://proceedings.mlr.press/v186/rodri-guez-lopez22a.html.

Related Material