通过特征重要性差值，解释模型更新为何导致性能变化。
Delta-Audit: Explaining What Changes When Models Change
- 用特征重要性差分 $Δϕ$ 定量分析模型版本间变化
- 深度提升的梯度提升模型在乳腺癌数据上变化最大（JSD≈0.357）
- 可识别真正影响行为的改动，适合模型审计与风险评估
模型更新（如超参数、网络深度、求解器或数据变更）常改变性能，但原因往往不透明。本文提出模型无关的 $Δ$-Attribution 框架，通过差分每特征重要性 $Δϕ(x)=ϕ_B(x)-ϕ_A(x)$ 来解释版本 $A$ 与 $B$ 之间的变化。我们构建了 $Δ$-Attribution 质量评估套件，涵盖幅度/稀疏性（L1、Top-$k$、熵）、一致性/偏移（秩重叠@10、JS 散度）、行为对齐（Delta Conservation Error, DCE；行为-归因耦合, BAC；CO$Δ$F）及鲁棒性（噪声、基线敏感性、分组遮蔽）。通过标准化空间中的快速遮蔽/钳制实现，采用类锚定边缘和基线平均。共审计 45 种设置：五类经典模型（逻辑回归、SVC、随机森林、梯度提升、kNN），三个数据集（乳腺癌、红酒、数字），每类三组 A/B 对比。结果发现：归纳偏置变更带来显著且行为一致的差异（如乳腺癌上 SVC 多项式→RBF：BAC≈0.998，DCE≈6.6；数字上随机森林特征规则互换：BAC≈0.997，DCE≈7.5），而“表面”调整（SVC gamma=scale vs auto，kNN 搜索方式）则秩重叠@10=1.0，DCE≈0。最显著的特征分配重分布出现在乳腺癌上的更深梯度提升模型（JS 散度≈0.357）。$Δ$-Attribution 提供轻量级更新审计能力，补充准确率，区分无害改动与行为显著或风险依赖变化。
原文摘要 · Abstract (English)
Model updates (new hyperparameters, kernels, depths, solvers, or data) change performance, but the \emph{reason} often remains opaque. We introduce \textbf{Delta-Attribution} (\mbox{$Δ$-Attribution}), a model-agnostic framework that explains \emph{what changed} between versions $A$ and $B$ by differencing per-feature attributions: $Δϕ(x)=ϕ_B(x)-ϕ_A(x)$. We evaluate $Δϕ$ with a \emph{$Δ$-Attribution Quality Suite} covering magnitude/sparsity (L1, Top-$k$, entropy), agreement/shift (rank-overlap@10, Jensen--Shannon divergence), behavioural alignment (Delta Conservation Error, DCE; Behaviour--Attribution Coupling, BAC; CO$Δ$F), and robustness (noise, baseline sensitivity, grouped occlusion). Instantiated via fast occlusion/clamping in standardized space with a class-anchored margin and baseline averaging, we audit 45 settings: five classical families (Logistic Regression, SVC, Random Forests, Gradient Boosting, $k$NN), three datasets (Breast Cancer, Wine, Digits), and three A/B pairs per family. \textbf{Findings.} Inductive-bias changes yield large, behaviour-aligned deltas (e.g., SVC poly$\!\rightarrow$rbf on Breast Cancer: BAC$\approx$0.998, DCE$\approx$6.6; Random Forest feature-rule swap on Digits: BAC$\approx$0.997, DCE$\approx$7.5), while ``cosmetic'' tweaks (SVC \texttt{gamma=scale} vs.\ \texttt{auto}, $k$NN search) show rank-overlap@10$=1.0$ and DCE$\approx$0. The largest redistribution appears for deeper GB on Breast Cancer (JSD$\approx$0.357). $Δ$-Attribution offers a lightweight update audit that complements accuracy by distinguishing benign changes from behaviourally meaningful or risky reliance shifts.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。