Reproduce “Graph-level Anomaly Detection via Hierarchical Memory Networks” 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
Chaoxi Niu, Guansong Pang, Ling Chen
HimNet introduces a hierarchical memory network for graph-level anomaly detection that learns normal patterns at both fine-grained node and holistic graph levels. The model achieved superior performance on 13 out of 16 real-world datasets and demonstrated robustness to anomaly contamination in training data.
Key insights
- Capturing hierarchical normal patterns (both fine-grained local and holistic global) is essential for comprehensive graph-level anomaly detection.
- Decoupling the reconstruction process from direct encoder outputs by using memory-approximated representations enhances the model's ability to differentiate anomalous graphs.
- Memory networks provide a robust mechanism to learn and store only "normal" graph prototypes, improving performance and stability in the presence of anomaly contamination.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.