Online Time Series Anomaly Detection with State Space Gaussian Processes
Citations Over Time
Abstract
We propose r-ssGPFA, an unsupervised online anomaly detection model for uni- and multivariate time series building on the efficient state space formulation of Gaussian processes. For high-dimensional time series, we propose an extension of Gaussian process factor analysis to identify the common latent processes of the time series, allowing us to detect anomalies efficiently in an interpretable manner. We gain explainability while speeding up computations by imposing an orthogonality constraint on the mapping from the latent to the observed. Our model's robustness is improved by using a simple heuristic to skip Kalman updates when encountering anomalous observations. We investigate the behaviour of our model on synthetic data and show on standard benchmark datasets that our method is competitive with state-of-the-art methods while being computationally cheaper.
Related Papers
- → Nearly Orthogonal Arrays With Mixed Levels and Small Runs(1992)42 cited
- → Discussion of some objections to thermomechanical orthogonality(1981)26 cited
- → Orthogonality in multiway classifications(1986)4 cited
- → Complements on Diminnie Orthogonality(1994)3 cited
- An Approximate Parameter Orthogonality(1998)