Reproduce “From Few-Shot to Zero-Shot: Towards Generalist Graph Anomaly Detection” 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
Yixin Liu, Shiyuan Li, Yu Zheng, Qingfeng Chen, Chengqi Zhang, Philip S. Yu, Shirui Pan
A generalist graph anomaly detection (GAD) framework is introduced, comprising ARC for few-shot and ARC_zero for zero-shot scenarios, designed to identify unusual nodes in diverse unseen graph datasets. The framework demonstrates superior anomaly detection performance across 17 real-world datasets, substantially lowering computational overhead and exhibiting robust generalization capabilities to novel graph structures and domains.
Key insights
- Effective generalist GAD requires aligning diverse feature spaces through smoothness-based sorting and capturing local, high-frequency structural differences indicative of anomalies.
- A cross-attentive in-context scoring mechanism can leverage a minimal number of labeled normal samples for reconstruction-based anomaly detection without requiring model optimization.
- Reliable zero-shot anomaly detection is feasible by iteratively identifying and refining pseudo-normal nodes to serve as context for anomaly scoring, even without any ground truth labels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.