HarsanyiNet: Computing Accurate Shapley Values in a Single Forward Propagation

Lu Chen, Siyu Lou, Keyan Zhang, Jin Huang, Quanshi Zhang
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:4804-4825, 2023.

Abstract

The Shapley value is widely regarded as a trustworthy attribution metric. However, when people use Shapley values to explain the attribution of input variables of a deep neural network (DNN), it usually requires a very high computational cost to approximate relatively accurate Shapley values in real-world applications. Therefore, we propose a novel network architecture, the HarsanyiNet, which makes inferences on the input sample and simultaneously computes the exact Shapley values of the input variables in a single forward propagation. The HarsanyiNet is designed on the theoretical foundation that the Shapley value can be reformulated as the redistribution of Harsanyi interactions encoded by the network.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-chen23s, title = {{H}arsanyi{N}et: Computing Accurate Shapley Values in a Single Forward Propagation}, author = {Chen, Lu and Lou, Siyu and Zhang, Keyan and Huang, Jin and Zhang, Quanshi}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {4804--4825}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/chen23s/chen23s.pdf}, url = {https://proceedings.mlr.press/v202/chen23s.html}, abstract = {The Shapley value is widely regarded as a trustworthy attribution metric. However, when people use Shapley values to explain the attribution of input variables of a deep neural network (DNN), it usually requires a very high computational cost to approximate relatively accurate Shapley values in real-world applications. Therefore, we propose a novel network architecture, the HarsanyiNet, which makes inferences on the input sample and simultaneously computes the exact Shapley values of the input variables in a single forward propagation. The HarsanyiNet is designed on the theoretical foundation that the Shapley value can be reformulated as the redistribution of Harsanyi interactions encoded by the network.} }
Endnote
%0 Conference Paper %T HarsanyiNet: Computing Accurate Shapley Values in a Single Forward Propagation %A Lu Chen %A Siyu Lou %A Keyan Zhang %A Jin Huang %A Quanshi Zhang %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-chen23s %I PMLR %P 4804--4825 %U https://proceedings.mlr.press/v202/chen23s.html %V 202 %X The Shapley value is widely regarded as a trustworthy attribution metric. However, when people use Shapley values to explain the attribution of input variables of a deep neural network (DNN), it usually requires a very high computational cost to approximate relatively accurate Shapley values in real-world applications. Therefore, we propose a novel network architecture, the HarsanyiNet, which makes inferences on the input sample and simultaneously computes the exact Shapley values of the input variables in a single forward propagation. The HarsanyiNet is designed on the theoretical foundation that the Shapley value can be reformulated as the redistribution of Harsanyi interactions encoded by the network.
APA
Chen, L., Lou, S., Zhang, K., Huang, J. & Zhang, Q.. (2023). HarsanyiNet: Computing Accurate Shapley Values in a Single Forward Propagation. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:4804-4825 Available from https://proceedings.mlr.press/v202/chen23s.html.

Related Material