AI Paper Atlas

Research map / Graph learning / GNN architectures

Watermarking Graph Neural Networks via Explanations for Ownership Protection

Jane Downer, Yingdan Shi, Ziyan Liu, Ren Wang, Binghui Wang

ICML 2026 · Graph learning › GNN architectures · approach: Robustness, privacy, and fairness of GNNs · 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
embed explanation watermark via dual-objective lossGraphSAGE · Photo, PubMed, CS test accuracytest accuracy: Photo 88.2 -> 90.8; PubMed 81.2 -> 86.0; CS 88.9 -> 88.4No clear effect
embed explanation watermark in Graph Transformer on Reddit2Graph Transformer · Reddit2 test accuracytest accuracy: 80.4 -> 79.4Hurt
embed explanation watermark (T=4 subgraphs)GNNs on Photo, PubMed, CS, Reddit2 · ownership verificationMI p-value: <0.001Improved
increase number of watermarked subgraphs T (ablation)explanation watermark · Photo, PubMed, CSalignment and p-valueImproved
increase watermarked subgraph size s (ablation)explanation watermark · Photo, PubMed, CSalignment, p-value, robustnessImproved
apply structured pruning attackGraphSAGE watermarked model · Photo, PubMed, CS, Reddit2MI p-valueNo clear effect
fine-tune at 0.1x learning rate for 49 epochsGraphSAGE watermarked model · Photo, PubMed, CS, Reddit2MI p-valueNo clear effect
fine-tune at 1x or 10x learning rateGraphSAGE watermarked model · fine-tuning attackwatermark presenceHurt
model merging of watermarked and fine-tuned modelswatermarked GCN/SAGE/Transformer · merged model verificationMI p-value: GCN 9.633e-7; SAGE 0.007; Transformer 0.008No clear effect
knowledge distillation attack without defensewatermarked GNN · removal attackMI p-value: p > 0.05Hurt
use GCN or SGC instead of GraphSAGE (ablation)watermarked model under fine-tuning and pruning · Photo, PubMed, CSrobustnessMixed
add watermarking to trainingGCN/SAGE training · training timewall time (s): Photo SAGE 109.06 -> 167.28Hurt

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