ESENSC:SHAP 的多项式时间公理化替代方案
ESENSC: A Polynomial-Time Axiomatic Alternative to SHAP
AI 导读
研究提出 ESENSC,一种高效且理论可靠的 SHAP 替代归因方法。它只需线性次数的联盟值评估即可给出确定性闭式归因,满足零玩家性质,并有基于效率、受限差分边际性和计算复杂度的公理化刻画。在神经网络与 XGBoost 表格预测任务上,ESENSC 与精确 SHAP 的归因量和特征排序高度一致,且计算成本远低于精确 SHAP,优于常用的采样近似方法。
正文 · 原文
Abstract:We propose ESENSC, a computationally efficient and theoretically grounded alternative to SHAP. ESENSC is a deterministic, closed-form attribution rule that requires only linearly many coalition-value evaluations, satisfies the null-player property, and admits an axiomatic characterization based on efficiency, a restricted form of differential marginality, and explicit computational this http URL, on tabular prediction tasks using neural network and XGBoost models, ESENSC shows close agreement with exact SHAP in both attribution magnitude and feature rankings while requiring substantially less computation. Under the evaluated settings, it compares favorably with widely used sampling-based SHAP approximations. While exact SHAP requires exponentially many coalition evaluations, ESENSC scales linearly with the number of features. These results show that ESENSC offers a favorable trade-off between agreement with exact SHAP and computational cost, providing a practical and theoretically grounded alternative for feature attribution in high-dimensional applications.
| Comments: | Accepted to NeurIPS 2026 (camera-ready version). 28 pages. Title changed; the method previously called ESENSC_rev2 is renamed ESENSC |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2603.00496 [cs.LG] |
| (or arXiv:2603.00496v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2603.00496 arXiv-issued DOI via DataCite |
Submission history
From: Junnosuke Shino Dr. [view email]
[v1]
Sat, 28 Feb 2026 06:30:35 UTC (462 KB)
[v2]
Fri, 13 Nov 2026 06:35:52 UTC (534 KB)
来源:arXiv cs.AI · arxiv.org