Reproduce “DP-DGAD: A Generalist Dynamic Graph Anomaly Detector with Dynamic Prototypes” 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
Jialun Zheng, Jie Liu, Jiannong Cao, Xiao Wang, Hanchen Yang, Yankai Chen, Philip S. Yu
DP-DGAD formulates dynamic graph anomaly detection as a cross-domain, label-free transfer problem. It combines temporal ego-graph representations with a memory of evolving normal and abnormal prototypes, then adapts to unlabeled target graphs using high-confidence pseudo-labels; across eight target datasets and three anomaly ratios, it achieved the best reported AUROC and AUPRC among the evaluated baselines.
Key insights
- Representing anomaly knowledge as a collection of dynamic prototypes is more flexible than relying on fixed class centers, helping preserve both domain-agnostic and domain-specific behavior.
- Confidence-based pseudo-label selection is more effective for target adaptation than random, threshold-based, distance-based, or similarity-based selection in the reported comparisons.
- Both source-domain diversity and adaptive scoring matter: removing domain adaptation or dynamic prototype scoring causes large performance losses.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.