.png){kind=icon}
Return
Generative adversarial message passing-based anomaly detection
.png){kind=icon}
OA
AI
DOI：10.1007/s44443-025-00021-6.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Anomaly detection plays a crucial role in ensuring the safety of mechanical equipment operations and the reliability of other critical systems in the real world. Existing generative adversarial nets (GANs)-based anomaly detection methods are limited to learning global distributions, lacking sensitivity to local features and relationships between objects, particularly performing poorly in sparse regions or detecting low-deviation local anomaly objects. To overcome this challenge, we propose a generative adversarial message passing-based (GAMP) method. First, the GANs is constrained to synthesize data that matches the normal objects in the original dataset under unsupervised conditions. Subsequently, an undirected graph is constructed using both the original and synthesized data. Message passing is then conducted among the nodes within this graph. Finally, anomaly objects are effectively identified based on the differences between objects. Experimental results on twelve publicly available datasets with diverse distributions demonstrate that the GAMP method outperforms the second-best competing algorithm, achieving a 10.7% increase in AUC. Moreover, the consistent performance across multiple public datasets highlights GAMP's broad application potential and practical significance in fields such as industrial automation, finance, and healthcare. By providing an effective solution for intelligent equipment monitoring and safety assurance, GAMP has the potential to enhance the reliability of critical systems across various industries.
Keywords:
Message passing
Generative adversarial nets
Anomaly detection
Outlier detection
Journal
J
IF:
6.1
Papers:
109
Citations:
9.5K
Organization
No organization information available
Cited Papers
.png){kind=icon}
ShareSave
ShareSaveShareSaveShareSave
View more