Research map / Time series / Anomaly detection
Multi-Scale Dynamic Graph Learning for Time Series Anomaly Detection (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MDGAD, a multi-scale dynamic graph structure learning approach with bilateral (forward/backward) forecasting errors for multivariate time series anomaly detection, evaluated on SWaT and WADI.
Problem
Multivariate time series anomaly detection with GNNs, where existing methods use static graphs and ignore time-varying and scale-dependent structure among time series.
Approach
Learns an evolving graph at each temporal scale (dilated inception TCN features feeding a scale-specific GRU-based graph learner with mix-skip propagation) and scores anomalies by combining forward and backward forecasting errors.
Key results
- MDGAD achieves F1 of 82.27 on SWaT versus 80.82 for GDN (best baseline).
- On WADI, MDGAD achieves F1 of 60.10 versus 56.92 for GDN, though GDN has higher precision (97.50 vs 82.47).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MDGAD instead of GDN | MTS anomaly detection · SWaT | F1: 80.82 -> 82.27 | Helped |
| Use MDGAD instead of GDN | MTS anomaly detection · WADI | F1: 56.92 -> 60.10 | Helped |
| Use MDGAD instead of GDN | MTS anomaly detection · WADI | Precision: 97.50 -> 82.47 | Hurt |
Surprising findings
- On WADI, GDN attains higher precision (97.50) than MDGAD (82.47), so MDGAD is not best on every metric.
Limitations
- Evaluated on only two datasets, and the paper is a short student abstract.
- Does not report ablations isolating the multi-scale or dynamic components.
- Precision on WADI is lower than GDN.
Experimental setup
Datasets: SWaT WADI
Benchmarks: SWaT WADI
Metrics: Precision Recall F1
Baselines: AE LSTM-VAE MAD-GAN GDN
Builds on: GDN MTAD-GAT
Closest papers by meaning
- Graph Mixture of Experts and Memory-augmented Routers for Multivariate Time Series Anomaly Detection (AAAI 2025 · similarity 0.89)
- GCAD: Anomaly Detection in Multivariate Time Series from the Perspective of Granger Causality (AAAI 2025 · similarity 0.88)
- A Generalizable Anomaly Detection Method in Dynamic Graphs (AAAI 2025 · similarity 0.87)
- The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark (NeurIPS 2024 · similarity 0.85)
- Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection (NeurIPS 2025 · similarity 0.85)
- Dynamic Spectral Graph Anomaly Detection (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.