Explaining Deep Tractable Probabilistic Models: The sum-product network case

Bhagirath Athresh Karanam, Saurabh Mathur, Predrag Radivojac, David M Haas, Kristian Kersting, Sriraam Natarajan
Proceedings of The 11th International Conference on Probabilistic Graphical Models, PMLR 186:325-336, 2022.

Abstract

We consider the problem of explaining a class of tractable deep probabilistic models, the Sum-Product Networks (SPNs) and present an algorithm ExSPN to generate explanations. To this effect, we define the notion of a context-specific independence tree(CSI-tree) and present an iterative algorithm that converts an SPN to a CSI-tree. The resulting CSI-tree is both interpretable and explainable to the domain expert. We achieve this by extracting the conditional independencies encoded by the SPN and approximating the local context specified by the structure of the SPN. Our extensive empirical evaluations on synthetic, standard, and real-world clinical data sets demonstrate that the CSI-tree exhibits superior explainability.

Cite this Paper

BibTeX
@InProceedings{pmlr-v186-karanam22a, title = {Explaining Deep Tractable Probabilistic Models: The sum-product network case}, author = {Karanam, Bhagirath Athresh and Mathur, Saurabh and Radivojac, Predrag and Haas, David M and Kersting, Kristian and Natarajan, Sriraam}, booktitle = {Proceedings of The 11th International Conference on Probabilistic Graphical Models}, pages = {325--336}, 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/karanam22a/karanam22a.pdf}, url = {https://proceedings.mlr.press/v186/karanam22a.html}, abstract = {We consider the problem of explaining a class of tractable deep probabilistic models, the Sum-Product Networks (SPNs) and present an algorithm ExSPN to generate explanations. To this effect, we define the notion of a context-specific independence tree(CSI-tree) and present an iterative algorithm that converts an SPN to a CSI-tree. The resulting CSI-tree is both interpretable and explainable to the domain expert. We achieve this by extracting the conditional independencies encoded by the SPN and approximating the local context specified by the structure of the SPN. Our extensive empirical evaluations on synthetic, standard, and real-world clinical data sets demonstrate that the CSI-tree exhibits superior explainability.} }
Endnote
%0 Conference Paper %T Explaining Deep Tractable Probabilistic Models: The sum-product network case %A Bhagirath Athresh Karanam %A Saurabh Mathur %A Predrag Radivojac %A David M Haas %A Kristian Kersting %A Sriraam Natarajan %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-karanam22a %I PMLR %P 325--336 %U https://proceedings.mlr.press/v186/karanam22a.html %V 186 %X We consider the problem of explaining a class of tractable deep probabilistic models, the Sum-Product Networks (SPNs) and present an algorithm ExSPN to generate explanations. To this effect, we define the notion of a context-specific independence tree(CSI-tree) and present an iterative algorithm that converts an SPN to a CSI-tree. The resulting CSI-tree is both interpretable and explainable to the domain expert. We achieve this by extracting the conditional independencies encoded by the SPN and approximating the local context specified by the structure of the SPN. Our extensive empirical evaluations on synthetic, standard, and real-world clinical data sets demonstrate that the CSI-tree exhibits superior explainability.
APA
Karanam, B.A., Mathur, S., Radivojac, P., Haas, D.M., Kersting, K. & Natarajan, S.. (2022). Explaining Deep Tractable Probabilistic Models: The sum-product network case. Proceedings of The 11th International Conference on Probabilistic Graphical Models, in Proceedings of Machine Learning Research 186:325-336 Available from https://proceedings.mlr.press/v186/karanam22a.html.

Related Material