arXiv cs.AI· Hyunseung Hwang, Seungeun Lee, Lucas Rosenblatt, Steven Euijong Whang, Julia Stoyanovich·· 9 天前AI 评分26
SHAP 解释的多重性：表征与评估
Explanation Multiplicity in SHAP: Characterization and Assessment
AI 导读
研究发现，即使模型、输入实例和预测结果固定不变，SHAP 解释在重复运行间也会出现显著分歧，作者将这一现象称为"解释多重性"。团队提出双种子协议区分模型引发与解释器引发的变异，并结合幅度、排名和集合三类指标及 Dirichlet、Mallows 零模型进行评估。结果显示该现象普遍存在，小数据集上模型引发的分歧更大，大数据集上解释器引发的分歧更大，CTE 等改进采样方法也无法消除排名层面的多重性。
正文 · 原文
Abstract:SHAP explanations are widely used in high-stakes settings to justify decisions, yet they can differ substantially across repeated runs, even when the model, the input instance, and the prediction are held fixed. Prior work has documented disagreement between explanation methods; we show that substantial disagreement arises even within SHAP across reruns of the same estimator on the same trained model and instance. We call this phenomenon explanation multiplicity and develop an evaluation methodology for characterizing it under deployment-realistic computational budgets, combining a dual-seed protocol that compares model-induced and explainer-induced variability, a hierarchy of magnitude-based, rank-based, and set-based metrics, and randomized Dirichlet and Mallows null models that provide reference scales for observed disagreement. Across multiple datasets, models, and sampling strategies, we find that explanation multiplicity is pervasive and persists even for high-confidence predictions. The relative contribution of each source depends on the data regime and model: model-induced disagreement is generally greater on smaller datasets, while explainer-induced disagreement is greater on larger datasets. Commonly used L2 distance can understate this instability, while rank-based metrics reveal substantial changes in top-ranked features, including the leading feature. Improved sampling methods such as CTE do not eliminate rank-level multiplicity, and K-Means reduces run-to-run variation while its compressed-background explanations can diverge from the empirical-distribution reference. Practitioners should treat single-run SHAP outputs as realizations of a distribution rather than as authoritative artifacts.
| Comments: | Accepted at NeurIPS 2026, Evaluations and Datasets Track |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2601.12654 [cs.LG] |
| (or arXiv:2601.12654v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2601.12654 arXiv-issued DOI via DataCite |
Submission history
From: Hyunseung Hwang [view email]
[v1]
Mon, 19 Jan 2026 02:01:18 UTC (668 KB)
[v2]
Sun, 25 Jan 2026 16:59:03 UTC (673 KB)
[v3]
Wed, 7 Oct 2026 20:48:57 UTC (2,090 KB)
来源：arXiv cs.AI · arxiv.org