SVARM-IQ: Efficient Approximation of Any-order Shapley Interactions through Stratification

Patrick Kolpaczki, Maximilian Muschalik, Fabian Fumagalli, Barbara Hammer, Eyke Hüllermeier
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:3520-3528, 2024.

Abstract

Addressing the limitations of individual attribution scores via the Shapley value (SV), the field of explainable AI (XAI) has recently explored intricate interactions of features or data points. In particular, extensions of the SV, such as the Shapley Interaction Index (SII), have been proposed as a measure to still benefit from the axiomatic basis of the SV. However, similar to the SV, their exact computation remains computationally prohibitive. Hence, we propose with SVARM-IQ a sampling-based approach to efficiently approximate Shapley-based interaction indices of any order. SVARM-IQ can be applied to a broad class of interaction indices, including the SII, by leveraging a novel stratified representation. We provide non-asymptotic theoretical guarantees on its approximation quality and empirically demonstrate that SVARM-IQ achieves state-of-the-art estimation results in practical XAI scenarios on different model classes and application domains.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-kolpaczki24a, title = { {SVARM-IQ}: Efficient Approximation of Any-order {S}hapley Interactions through Stratification }, author = {Kolpaczki, Patrick and Muschalik, Maximilian and Fumagalli, Fabian and Hammer, Barbara and H\"{u}llermeier, Eyke}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {3520--3528}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/kolpaczki24a/kolpaczki24a.pdf}, url = {https://proceedings.mlr.press/v238/kolpaczki24a.html}, abstract = { Addressing the limitations of individual attribution scores via the Shapley value (SV), the field of explainable AI (XAI) has recently explored intricate interactions of features or data points. In particular, extensions of the SV, such as the Shapley Interaction Index (SII), have been proposed as a measure to still benefit from the axiomatic basis of the SV. However, similar to the SV, their exact computation remains computationally prohibitive. Hence, we propose with SVARM-IQ a sampling-based approach to efficiently approximate Shapley-based interaction indices of any order. SVARM-IQ can be applied to a broad class of interaction indices, including the SII, by leveraging a novel stratified representation. We provide non-asymptotic theoretical guarantees on its approximation quality and empirically demonstrate that SVARM-IQ achieves state-of-the-art estimation results in practical XAI scenarios on different model classes and application domains. } }
Endnote
%0 Conference Paper %T SVARM-IQ: Efficient Approximation of Any-order Shapley Interactions through Stratification %A Patrick Kolpaczki %A Maximilian Muschalik %A Fabian Fumagalli %A Barbara Hammer %A Eyke Hüllermeier %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-kolpaczki24a %I PMLR %P 3520--3528 %U https://proceedings.mlr.press/v238/kolpaczki24a.html %V 238 %X Addressing the limitations of individual attribution scores via the Shapley value (SV), the field of explainable AI (XAI) has recently explored intricate interactions of features or data points. In particular, extensions of the SV, such as the Shapley Interaction Index (SII), have been proposed as a measure to still benefit from the axiomatic basis of the SV. However, similar to the SV, their exact computation remains computationally prohibitive. Hence, we propose with SVARM-IQ a sampling-based approach to efficiently approximate Shapley-based interaction indices of any order. SVARM-IQ can be applied to a broad class of interaction indices, including the SII, by leveraging a novel stratified representation. We provide non-asymptotic theoretical guarantees on its approximation quality and empirically demonstrate that SVARM-IQ achieves state-of-the-art estimation results in practical XAI scenarios on different model classes and application domains.
APA
Kolpaczki, P., Muschalik, M., Fumagalli, F., Hammer, B. & Hüllermeier, E.. (2024). SVARM-IQ: Efficient Approximation of Any-order Shapley Interactions through Stratification . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:3520-3528 Available from https://proceedings.mlr.press/v238/kolpaczki24a.html.

Related Material