Flow-based Attribution in Graphical Models: A Recursive Shapley Approach

Raghav Singal, George Michailidis, Hoiyi Ng
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:9733-9743, 2021.

Abstract

We study the attribution problem in a graphical model, wherein the objective is to quantify how the effect of changes at the source nodes propagates through the graph. We develop a model-agnostic flow-based attribution method, called recursive Shapley value (RSV). RSV generalizes a number of existing node-based methods and uniquely satisfies a set of flow-based axioms. In addition to admitting a natural characterization for linear models and facilitating mediation analysis for non-linear models, RSV satisfies a mix of desirable properties discussed in the recent literature, including implementation invariance, sensitivity, monotonicity, and affine scale invariance.

Cite this Paper


BibTeX
@InProceedings{pmlr-v139-singal21a, title = {Flow-based Attribution in Graphical Models: A Recursive Shapley Approach}, author = {Singal, Raghav and Michailidis, George and Ng, Hoiyi}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {9733--9743}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/singal21a/singal21a.pdf}, url = {https://proceedings.mlr.press/v139/singal21a.html}, abstract = {We study the attribution problem in a graphical model, wherein the objective is to quantify how the effect of changes at the source nodes propagates through the graph. We develop a model-agnostic flow-based attribution method, called recursive Shapley value (RSV). RSV generalizes a number of existing node-based methods and uniquely satisfies a set of flow-based axioms. In addition to admitting a natural characterization for linear models and facilitating mediation analysis for non-linear models, RSV satisfies a mix of desirable properties discussed in the recent literature, including implementation invariance, sensitivity, monotonicity, and affine scale invariance.} }
Endnote
%0 Conference Paper %T Flow-based Attribution in Graphical Models: A Recursive Shapley Approach %A Raghav Singal %A George Michailidis %A Hoiyi Ng %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-singal21a %I PMLR %P 9733--9743 %U https://proceedings.mlr.press/v139/singal21a.html %V 139 %X We study the attribution problem in a graphical model, wherein the objective is to quantify how the effect of changes at the source nodes propagates through the graph. We develop a model-agnostic flow-based attribution method, called recursive Shapley value (RSV). RSV generalizes a number of existing node-based methods and uniquely satisfies a set of flow-based axioms. In addition to admitting a natural characterization for linear models and facilitating mediation analysis for non-linear models, RSV satisfies a mix of desirable properties discussed in the recent literature, including implementation invariance, sensitivity, monotonicity, and affine scale invariance.
APA
Singal, R., Michailidis, G. & Ng, H.. (2021). Flow-based Attribution in Graphical Models: A Recursive Shapley Approach. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:9733-9743 Available from https://proceedings.mlr.press/v139/singal21a.html.

Related Material