CLOSURE detects time-series anomalies via consistency and recoverability
Original title: CLOSURE: Cross-Channel Closure for Multivariate Time Series Anomaly Detection
A diffusion reconstructor and a relation-closure operator are trained separately and calibrated only on normal data.
CLOSURE treats multivariate anomaly detection as two separate normal-data prediction tasks: recoverability and cross-channel consistency. A diffusion reconstructor conditioned on a discrete memory of normal temporal bases handles recovery, while a relation-closure operator predicts each channel only from other channels using sparse signed gates. Scores are reconciled through calibration on normal data. On public benchmark streams, the authors report that CLOSURE outperforms recent reconstruction-, frequency-, and topology-based detectors.
- The method targets unlabeled monitoring of industrial systems and server clusters.
- A residual alone can miss a sensor that drifts from its peers.
- The target channel's own trajectory is withheld from the consistency predictor.
From the original article
1 min read at arxiv.orgMultivariate time series produced by industrial installations and server clusters are under continuous monitoring, and anomaly detection is the first line of defense before a local fault escalates into a system-wide failure. In the absence of labels, the dominant paradigm trains a reconstruction model on normal data and scores the residual to reach a verdict.
By Fan Xu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.