ORTHOGONAL PERMUTATION SAMPLING FOR SHAPLEY VALUES: UNBIASED STRATIFIED ESTIMATORS WITH VARIANCE GUARANTEES
Journal of Intelligent Systems and Applied Data Science · 2026
DOI: 10.64680/jisads.v3i2.44 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Intelligent Systems and Applied Data Science, 2026. Verfasst von YASH VARSHNEY und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Shapley values for feature attribution suffer from high variance requiring thousands of model evaluations. We introduce Orthogonal Permutation Sampling (OPS), achieving provable variance reduction through: (i) exact position stratification, (ii) antithetic permutation coupling, and (iii) control variates. We prove finite-sample variance dominance over Monte Carlo and non-positive covariance under submodularity. Empirical validation across six benchmarks shows 5-26× variance reduction for typical dimensions (n=10-20) and 67× for n=50. OPS achieves 2-5× lower MSE than KernelSHAP at equivalent budgets with 7% runtime overhead (all p
Zitieren
APA
VARSHNEY, Y., TYAGI, R., & SINHA, A. (2026). ORTHOGONAL PERMUTATION SAMPLING FOR SHAPLEY VALUES: UNBIASED STRATIFIED ESTIMATORS WITH VARIANCE GUARANTEES. Journal of Intelligent Systems and Applied Data Science. https://doi.org/10.64680/jisads.v3i2.44
Vancouver
VARSHNEY Y, TYAGI R, SINHA A. ORTHOGONAL PERMUTATION SAMPLING FOR SHAPLEY VALUES: UNBIASED STRATIFIED ESTIMATORS WITH VARIANCE GUARANTEES. Journal of Intelligent Systems and Applied Data Science. 2026. doi:10.64680/jisads.v3i2.44
BibTeX
@article{VARSHNEY2026,
author = {VARSHNEY, YASH and TYAGI, RANAV and SINHA, ANURAG},
title = {ORTHOGONAL PERMUTATION SAMPLING FOR SHAPLEY VALUES: UNBIASED STRATIFIED ESTIMATORS WITH VARIANCE GUARANTEES},
journal = {Journal of Intelligent Systems and Applied Data Science},
year = {2026},
doi = {10.64680/jisads.v3i2.44}
}
EUVIMED