Reproduce “Feature-Context Consistency: Unsupervised Adversarial Detection with Drift Stability on Attributed Graphs” 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
Firas Bayram, Maxime Cordy
CB-RBM detects graph anomalies by measuring how much neighborhood context improves its explanation of node features, without anomaly training labels. On XBlock and Reddit, it achieves AUROC of at least 0.96 across four injected attack types and average Combined Scores of at least 0.87 under simulated 20-step drift. The signal weakens when attackers gain genuine connections or feature–context correlation is absent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.