Reproduce “An Attention-Based Deep Generative Model for Anomaly Detection in Industrial Control Systems” 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
Mayra Macas, Chunming Wu, Walter Fuertes
Researchers from Zhejiang University and Universidad de las Fuerzas Armadas ESPE developed an attention-based Convolutional Variational Autoencoder (aCVAE) for unsupervised anomaly detection in Industrial Control Systems. The model achieves an F1 score of 89% and high recall on the SWaT testbed, demonstrating improved stability and reduced false positives by employing reconstruction probability and dynamic thresholding.
Key insights
- 3D convolutional layers are highly effective at simultaneously capturing both spatial correlations between sensors and temporal dependencies across different observation windows, outperforming 2D convolutions for ICS data.
- Integrating a 3D-Convolutional Block Attention Module significantly enhances feature representation and anomaly detection accuracy by enabling the model to focus on the most relevant information within the complex correlation matrices.
- Using reconstruction probability from a Variational Autoencoder provides a more robust and probabilistic measure of anomaly that inherently accounts for data variability, leading to better distinction of subtle anomalies compared to deterministic reconstruction error.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.