SHAP-IQ: Unified Approximation of any-order Shapley Interactions

Fumagalli F, Muschalik M, Kolpaczki P, Hüllermeier E, Hammer B (2023)
In: Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Advances in Neural Information Processing Systems. .

Konferenzbeitrag | Veröffentlicht | Englisch
 
Download
Es wurden keine Dateien hochgeladen. Nur Publikationsnachweis!
Autor*in
Fumagalli, FabianUniBi ; Muschalik, Maximilian; Kolpaczki, Patrick; Hüllermeier, Eyke; Hammer, BarbaraUniBi
Abstract / Bemerkung
Predominately in explainable artificial intelligence (XAI) research, the Shapley value (SV) is applied to determine feature attributions for any black box model. Shapley interaction indices extend the SV to define any-order feature interactions. Defining a unique Shapley interaction index is an open research question and, so far, three definitions have been proposed, which differ by their choice of axioms. Moreover, each definition requires a specific approximation technique. Here, we propose SHAPley Interaction Quantification (SHAP-IQ), an efficient sampling-based approximator to compute Shapley interactions for arbitrary cardinal interaction indices (CII), i.e. interaction indices that satisfy the linearity, symmetry and dummy axiom. SHAP-IQ is based on a novel representation and, in contrast to existing methods, we provide theoretical guarantees for its approximation quality, as well as estimates for the variance of the point estimates. For the special case of SV, our approach reveals a novel representation of the SV and corresponds to Unbiased KernelSHAP with a greatly simplified calculation. We illustrate the computational efficiency and effectiveness by explaining language, image classification and high-dimensional synthetic models.
Erscheinungsjahr
2023
Titel des Konferenzbandes
Advances in Neural Information Processing Systems 36 (NeurIPS 2023)
Serien- oder Zeitschriftentitel
Advances in Neural Information Processing Systems
Konferenz
37th Conference on Neural Information Processing Systems (NeurIPS)
Konferenzort
New Orleans, LA
Konferenzdatum
2023-12-10 – 2023-12-16
Page URI
https://pub.uni-bielefeld.de/record/2987580

Zitieren

Fumagalli F, Muschalik M, Kolpaczki P, Hüllermeier E, Hammer B. SHAP-IQ: Unified Approximation of any-order Shapley Interactions. In: Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Advances in Neural Information Processing Systems. 2023.
Fumagalli, F., Muschalik, M., Kolpaczki, P., Hüllermeier, E., & Hammer, B. (2023). SHAP-IQ: Unified Approximation of any-order Shapley Interactions. Advances in Neural Information Processing Systems 36 (NeurIPS 2023), Advances in Neural Information Processing Systems
Fumagalli, Fabian, Muschalik, Maximilian, Kolpaczki, Patrick, Hüllermeier, Eyke, and Hammer, Barbara. 2023. “SHAP-IQ: Unified Approximation of any-order Shapley Interactions”. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Advances in Neural Information Processing Systems.
Fumagalli, F., Muschalik, M., Kolpaczki, P., Hüllermeier, E., and Hammer, B. (2023). “SHAP-IQ: Unified Approximation of any-order Shapley Interactions” in Advances in Neural Information Processing Systems 36 (NeurIPS 2023) Advances in Neural Information Processing Systems.
Fumagalli, F., et al., 2023. SHAP-IQ: Unified Approximation of any-order Shapley Interactions. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Advances in Neural Information Processing Systems.
F. Fumagalli, et al., “SHAP-IQ: Unified Approximation of any-order Shapley Interactions”, Advances in Neural Information Processing Systems 36 (NeurIPS 2023), Advances in Neural Information Processing Systems, 2023.
Fumagalli, F., Muschalik, M., Kolpaczki, P., Hüllermeier, E., Hammer, B.: SHAP-IQ: Unified Approximation of any-order Shapley Interactions. Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Advances in Neural Information Processing Systems. (2023).
Fumagalli, Fabian, Muschalik, Maximilian, Kolpaczki, Patrick, Hüllermeier, Eyke, and Hammer, Barbara. “SHAP-IQ: Unified Approximation of any-order Shapley Interactions”. Advances in Neural Information Processing Systems 36 (NeurIPS 2023). 2023. Advances in Neural Information Processing Systems.

Link(s) zu Volltext(en)
Access Level
Restricted Closed Access

Export

Markieren/ Markierung löschen
Markierte Publikationen

Open Data PUB

Web of Science

Dieser Datensatz im Web of Science®
Quellen

arXiv: 2303.01179

Suchen in

Google Scholar