Contrast to Detect: Dynamic Graph Contrastive Regularization for Unsupervised Anomaly Detection in Multivariate Time Series
arXiv:2605.23744v1 Announce Type: new Abstract: Anomaly detection in multivariate time series (MTS) is hindered by dynamic inter-variable dependencies and feature entanglement under spectral noise, and in practice, is further complicated by the absence of anomaly labels. Existing reconstruction-based detectors tend to recover anomalies as faithfully as normal patterns, while prevailing graph contrastive methods enforce invariance across views and thus assume a stationary relational structure, an assumption that breaks under structural drift in real systems. We propose ContrastAD, an unsupervis
The proliferation of complex, multivariate time series data from various systems necessitates more robust unsupervised anomaly detection methods, especially as AI applications become more critical.
Improved unsupervised anomaly detection, particularly in complex systems, enhances reliability, security, and efficiency across numerous applications from industrial control to predictive maintenance.
This research introduces a new approach to anomaly detection that better handles dynamic inter-variable dependencies and structural drift, which are common challenges in real-world data.
- · AI/ML researchers
- · Industrial IoT platforms
- · Cybersecurity firms
- · Predictive maintenance providers
- · Systems reliant on outdated anomaly detection methods
- · Industries with high false positive rates in anomaly detection
More accurate and reliable identification of system failures, intrusions, and operational inefficiencies.
Reduced operational costs and enhanced security for critical infrastructure and complex technical systems.
Accelerated deployment of autonomous systems and agents by increasing their resilience to unforeseen events and sensor noise.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG