Reproduce “Certifying Graph Neural Networks Against Label and Structure Poisoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lukas Gosch, Xichuan Chen, Yan Scholten, Stephan Günnemann
ST-GPA certifies node predictions against label and edge poisoning by enriching partitions with predicted labels, links, or both before training classifiers and voting. This addresses sparse supervision and connectivity left by naïve partitioning. Experiments report higher certified accuracy across label, structure, and combined attacks than naïve partitioning; the joint result is on PubMed. Link generation assumes homophily and adds cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.