On Measuring Intrinsic Causal Attributions in Deep Neural Networks

Saptarshi Saha, Dhruv Vansraj Rathore, Soumadeep Saha, David Doermann, Utpal Garain
Proceedings of the Fourth Conference on Causal Learning and Reasoning, PMLR 275:1405-1434, 2025.

Abstract

Quantifying the causal influence of input features within neural networks has become a topic of increasing interest. Existing approaches typically assess direct, indirect, and total causal effects. This work treats NNs as structural causal models (SCMs) and extends our focus to include intrinsic causal contributions (ICC). We propose an identifiable generative post-hoc framework for quantifying ICC. We also draw a relationship between ICC and Sobol’ indices. Our experiments on synthetic and real-world datasets demonstrate that ICC generates more intuitive and faithful explanations compared to existing global explanation techniques.

Cite this Paper

BibTeX
@InProceedings{pmlr-v275-saha25a, title = {On Measuring Intrinsic Causal Attributions in Deep Neural Networks}, author = {Saha, Saptarshi and Rathore, Dhruv Vansraj and Saha, Soumadeep and Doermann, David and Garain, Utpal}, booktitle = {Proceedings of the Fourth Conference on Causal Learning and Reasoning}, pages = {1405--1434}, year = {2025}, editor = {Huang, Biwei and Drton, Mathias}, volume = {275}, series = {Proceedings of Machine Learning Research}, month = {07--09 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v275/main/assets/saha25a/saha25a.pdf}, url = {https://proceedings.mlr.press/v275/saha25a.html}, abstract = {Quantifying the causal influence of input features within neural networks has become a topic of increasing interest. Existing approaches typically assess direct, indirect, and total causal effects. This work treats NNs as structural causal models (SCMs) and extends our focus to include intrinsic causal contributions (ICC). We propose an identifiable generative post-hoc framework for quantifying ICC. We also draw a relationship between ICC and Sobol’ indices. Our experiments on synthetic and real-world datasets demonstrate that ICC generates more intuitive and faithful explanations compared to existing global explanation techniques.} }
Endnote
%0 Conference Paper %T On Measuring Intrinsic Causal Attributions in Deep Neural Networks %A Saptarshi Saha %A Dhruv Vansraj Rathore %A Soumadeep Saha %A David Doermann %A Utpal Garain %B Proceedings of the Fourth Conference on Causal Learning and Reasoning %C Proceedings of Machine Learning Research %D 2025 %E Biwei Huang %E Mathias Drton %F pmlr-v275-saha25a %I PMLR %P 1405--1434 %U https://proceedings.mlr.press/v275/saha25a.html %V 275 %X Quantifying the causal influence of input features within neural networks has become a topic of increasing interest. Existing approaches typically assess direct, indirect, and total causal effects. This work treats NNs as structural causal models (SCMs) and extends our focus to include intrinsic causal contributions (ICC). We propose an identifiable generative post-hoc framework for quantifying ICC. We also draw a relationship between ICC and Sobol’ indices. Our experiments on synthetic and real-world datasets demonstrate that ICC generates more intuitive and faithful explanations compared to existing global explanation techniques.
APA
Saha, S., Rathore, D.V., Saha, S., Doermann, D. & Garain, U.. (2025). On Measuring Intrinsic Causal Attributions in Deep Neural Networks. Proceedings of the Fourth Conference on Causal Learning and Reasoning, in Proceedings of Machine Learning Research 275:1405-1434 Available from https://proceedings.mlr.press/v275/saha25a.html.

Related Material