International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Delta-Audit: Explaining What Changes When Models Change

Authors :

Arshia Hemmat¹ Afsane Fatemi²

1- university of isfahan 2- Department of Software Engineering Univesity of Isfahan

Keywords :

Explainable AI (XAI)،feature attribution،delta attribution،robustness،distribution shift

Abstract :

Model updates (new hyperparameters, kernels, depths, solvers, or data) change performance, but the \emph{reason} often remains opaque. We introduce \textbf{Delta-Attribution} (\mbox{$\Delta$-Attribution}), a model-agnostic framework that explains \emph{what changed} between versions $A$ and $B$ by differencing per-feature attributions: $\Delta\phi(x)=\phi_B(x)-\phi_A(x)$. We evaluate $\Delta\phi$ with a \emph{$\Delta$-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$\Delta$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). $\Delta$-Attribution offers a lightweight update audit that complements accuracy by distinguishing benign changes from behaviourally meaningful or risky reliance shifts.

Papers List

List of archived papers

Intelligent Resource Collision Management for Cellular Vehicular Systems Using Software-Defined Networking

Mohammad Kazemiesfeh - Neda Moghim - Ahmadreza Montazerolghaem

Word-level Persian Lipreading Dataset

Javad Peymanfard - Ali Lashini - Samin Heydarian - Hossein Zeinali - Nasser Mozayani

Density Estimation Helps Adversarial Robustness

Afsaneh Hasanebrahimi - Bahareh Kaviani Baghbaderani - Reshad Hosseini - Ahmad Kalhor

Area-Efficient VLSI Implementation of Bit-Serial Multiplier Using Polynomial Basis over GF(2m)

Saeideh Nabipour - Javad Javidan - Gholamreza Zare Fatin

Deep Learning-Based Malaysian Sign Language (MSL) Recognition: Exploring the Impact of Color Spaces

Ervin Gubin Moung - Precilla Fiona Suwek - Maisarah Mohd Sufian - Valentino Liaw - Ali Farzamnia - Wei Leong Khong

Automated Person Identification from Hand Images\\using Hierarchical Vision Transformer Network

Zahra Ebrahimian - Seyed Ali Mirsharji - Ramin Toosi - Mohammad Ali Akhaee

A Comprehensive Dataset of Real-scene Images for Text Detection and Recognition in Persian

Iman Souzanchi - Ramin Rahimi - Mohammad Ali Majidi Anvari - Atefeh Baniasadi - Ashkan Sadeghi - Mohammad Reza Mohammadi

Exploring 3D Transfer Learning CNN Models for Alzheimer’s Disease Diagnosis from MRI Images

Fatemehsadat Ghanadi Ladani - Hamidreza Baradaran Kashani

An influence maximization algorithm based on community detection using topological features

Zahra Aghaee - Afsaneh Fatemi

Real-time Implementation of Fuzzy Visual Servoing for a Delta Robot via Shape and Color Detection

Nooshin Najafian - Alireza Ashrafi Majd - Abbas Ansaroudi - Sahar Aghazadeh - Manizheh Zakeri - Mohammad-Reza Sayyed Noorani