AI Paper Atlas

Research map / Graph learning / GNN architectures

Generating In-Distribution Counterfactual Explanation for Graph Neural Networks

Linmao Chen, Chaobo He, Junwei Cheng, Chunying Li, Quanlong Guan

AAAI 2026 · Graph learning › GNN architectures · approach: GNN explainability · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use ICExplainer (diffusion with variational prior) instead of D4Explainercounterfactual GNN explainers · Tree-Circles, Tree-Grid, BA-2motifs, MUTAGvalidity: 0.905 -> 0.975 (Tree-Circles); 0.789 -> 0.937 (BA-2motifs)Improved
use ICExplainer instead of leading baselinecounterfactual GNN explainers · four datasetssparsity: 0.104 -> 0.049 (MUTAG); 0.096 (ProxyExplainer) vs 0.107 (ICE) on BA-2motifsMixed
use ICExplainer instead of D4Explainercounterfactual GNN explainers · Tree-Circles, BA-2motifs, MUTAGfidelity: 0.854 -> 0.913; 0.727 -> 0.843; 0.537 -> 0.630Improved
use ICExplainer instead of D4Explainercounterfactual GNN explainers · Tree-Circles, Tree-Grid, BA-2motifs, MUTAGMMD sum to original graphs: 0.1400 -> 0.0844; 0.1011 vs 0.0812; ProxyExplainer 0.0996 vs ICE 0.0812 on BA-2motifsMixed
use GCN encoder instead of SAGE (ablation)ICExplainer · four datasetsvalidity: 0.961 -> 0.975 (Tree-Circles); 0.867 -> 0.937 (BA-2motifs)Improved
use GCN encoder instead of GIN (ablation)ICExplainer · four datasetsvalidity: 0.782 -> 0.843 (MUTAG); 0.954 -> 0.975 (Tree-Circles)Improved
increase prediction loss weight lambda (ablation)ICExplainer · Tree-Circles vs BA-2motifsvalidity/fidelityMixed
increase number of PPGN layers L from 3 to 8 (ablation)ICExplainer · Tree-Circles, BA-2motifsperformanceImproved
increase latent dimension D from 8 to 128 (ablation)ICExplainer · four datasetsperformanceNo clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy