Signature-Kernel One-Class Learning with Manifold Regularisation for Multivariate Time Series Anomaly Detection

Name: Signature-Kernel One-Class Learning with Manifold Regularisation for Multivariate Time Series Anomaly Detection
Uploaded: 2026-10-15T11:29:27.817440+00:00
Duration: 825 s
Description: The method generalizes large-margin $\ell_p$-SVDD for time series anomaly detection by adding manifold regularization. This regularization term enforces smoothness on the data's underlying structure, capturing regularities in normal behavior. Theoretical analysis shows this improves generalization bounds, and experiments demonstrate superior performance on various time series datasets.

The method generalizes large-margin $\ell_p$-SVDD for time series anomaly detection by adding manifold regularization. This regularization term enforces smoothness on the data's underlying structure, capturing regularities in normal behavior. Theoretical analysis shows this improves generalization bounds, and experiments demonstrate superior performance on various time series datasets.

Open the full episode