New unsupervised method detects graph attacks despite data drift

Researchers have developed a novel unsupervised method called Feature-Context Consistency (CB-RBM) to detect adversarial attacks on evolving attributed graphs. This approach leverages the alignment between a node's attributes and its neighborhood's aggregated attributes to maintain detection accuracy even when fraud labels are unavailable and the data distribution naturally shifts over time. CB-RBM demonstrated strong performance, achieving an AUROC of at least 0.96 on four adversarial attacks and maintaining a low false positive rate under drift, outperforming supervised graph neural networks and other unsupervised baselines. AI

IMPACT This unsupervised approach could improve the robustness of graph-based security systems against evolving threats and data drift.

RANK_REASON The cluster contains a research paper detailing a new method for detecting adversarial attacks on attributed graphs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• CB-RBM
• Feature-Context Consistency
• graph neural network
• Reddit
• restricted Boltzmann machine
• XBlock Ethereum

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →