Periodic Unitary Encoding for Quantum Anomaly Detection of Temporal Series

Daniele Lizzio Bosco, Riccardo Romanello, Giuseppe Serra

Abstract: Anomaly detection in temporal series is a compelling area of research with applications in fields such as finance, healthcare, and predictive maintenance. Recently, Quantum Machine Learning (QML) has emerged as a promising approach to tackle such problems, leveraging the unique properties of quantum systems. Among QML techniques, kernel-based methods have gained significant attention due to their ability to effectively handle both supervised and unsupervised tasks. In the context of anomaly detection, unsupervised approaches are particularly valuable as labeled data is often scarce. Nevertheless, temporal series data frequently exhibit known seasonality, even in unsupervised settings. We propose a novel quantum kernel designed to incorporate seasonality information into anomaly detection tasks. Our approach constructs a Hamiltonian matrix that induces a unitary operator which period corresponds to the seasonality of the task under consideration. This unitary operator is then used to

External IDs:dblp:conf/iqsoft/BoscoR025